Machine Learning Prediction Model for Durability of Geopolymer Concrete with CDW and Artificial Aggregates under Harsh Environmental Exposure

Machine Learning Prediction Model for Durability of Geopolymer Concrete with CDW and Artificial Aggregates under Harsh Environmental Exposure | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Machine Learning Prediction Model for Durability of Geopolymer Concrete with CDW and Artificial Aggregates under Harsh Environmental Exposure Atul S. Kurzekar, Uday P. Waghe, Prajakta Waghe This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6862865/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 04 Aug, 2025 Read the published version in Iranian Journal of Science and Technology, Transactions of Civil Engineering → Version 1 posted 12 You are reading this latest preprint version Abstract This research explores how machine learning (ML) models can predict the durability of geopolymer concrete (GPC) produced using construction and demolition waste (CDW) as fine aggregate and artificial lightweight coarse aggregates. Durability was assessed under harsh exposure conditions, including sulphate attack, chloride penetration, and freeze–thaw cycles. The study involved 90 experimental samples across varying CDW replacement levels. Five regression models, such as Linear Regression, Support Vector Regression, Random Forest, XGBoost, and Artificial Neural Networks, were trained to forecast strength retention based on mix composition and environmental exposure. Among them, XGBoost demonstrated the highest predictive accuracy (R² = 0.96). SHAP analysis was used to explain model predictions and identify key influencing parameters, with CDW content and the activator-to-binder ratio emerging as critical factors. The findings show that moderate CDW incorporation enhances durability while reducing environmental impact, and that interpretable ML tools can assist in optimising mix designs for long-term performance in aggressive environments. Geopolymer concrete Artificial aggregate Machine Learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 1. Introduction The construction sector today faces growing pressure to reduce its environmental impact and manage natural resources more responsibly. Traditional concrete, which relies heavily on Portland cement and natural aggregates, contributes significantly to carbon emissions and the depletion of raw materials [ 1 ]. As a response, geopolymer concrete (GPC) has been proposed as a more sustainable alternative [ 2 ]. It offers lower carbon emissions and the ability to incorporate waste materials from other industries. Among these, various types of waste considered for use as a natural fine aggregate and used in concrete mixtures, CDW has received the most attention [ 3 ]. CDW containing crushed bricks may help improve the internal structure of the concrete and provide some pozzolanic activity. Sulphate attack, chloride attack and freeze-thaw cycles are key conditions to consider when planning for long-term durability [ 4 ]. At almost the same moment, machine-learning software began to pop up in laboratory notebooks, advertising a chance to snap quick forecasts out of wobbly test data. Open a dashboard, feed in a few dozen measurements, and the models sift through the hidden connections, slashing that brutal grind of trial-and-error pouring. In concrete circles, engineers now lean on those calculations to pin down durability and tinker with mixture recipes in a fraction of the usual fuss [ 5 ] Concrete researchers have lately turned to machine learning to forecast how mixes perform, with geopolymer concrete drawing particular interest. Early experiments leaned on familiar tools such as linear regression or support-vector regression, but the field has quickly pivoted toward ensemble giants like random forest and XGBoost [ 6 ]. Those algorithms tackle the tangled, non-linear relationships in the data with far less fuss. Artificial Neural Networks (ANNs) have also shown promise in capturing intricate mix-performance relationships, albeit with trade-offs in interpretability. However, most prior studies focus on strength prediction alone and seldom integrate explainable AI techniques such as SHAP to assess feature importance under varied durability exposures. To address this gap, the present study evaluates five diverse ML models—LR, SVR, RF, ANN, and XGBoost—to predict compressive strength retention in CDW-based GPC exposed to harsh environments. The selection of models reflects a deliberate mix of linear, kernel-based, tree-based, and neural network methods to assess predictive performance and model transparency under different algorithmic paradigms [ 7 ]. This research seeks to assess the durability performance of CDW and artificial aggregate-based geopolymer concrete with different contents under severe exposure conditions. It also assesses the predictive capacity of different ML algorithms to compressive strength retention, with feature importance analysis based on SHAP values [ 8 ]. The experimental and computational strategy seeks to inform the design of sustainable, durable GPC products that can perform under actual environmental conditions. The novelty of this study is that it has an integrated approach to addressing a critical research issue in GPC durability. Some studies have investigated the mechanical properties of GPC or addressed some environmental conditions, but limited research has considered the integration of recycled CDW fine aggregate and artificial coarse aggregates and with a full durability test under severe conditions such as sulphate attack, freeze–thaw, and chloride penetration. This work fills this gap by presenting comprehensive experimental data on how these parameters interact with each other and further employs machine learning models, namely XGBoost with SHAP analysis, to predict durability and emphasise how each mix parameter influences it. The combination of experimental and predictive modelling approaches offers a feasible and innovative solution to the creation of more durable and sustainable GPC mixes. This work presents a new framework that facilitates the use of recycled materials in real-world applications, to the benefit of the environment and performance under harsh service conditions. 2. Literature Review Geopolymer concrete (GPC) is being studied as a more environmentally friendly option compared to ordinary Portland cement (OPC), mainly because it can make use of industrial waste materials and helps reduce carbon emissions [ 9 ]. Materials like fly ash and ground glass powder are commonly used as sources of aluminosilicate, which, when mixed with alkaline solutions, react to form a dense and durable binder. Research has shown that adding glass powder can improve the concrete’s structure and increase its resistance to chemical attacks such as sulphate and chloride exposure [ 10 ]. At the same time, interest in recycling has grown, especially with the push toward circular economy practices. This has led to more studies exploring how construction and demolition waste (CDW) can be reused in concrete production [ 11 ]. CDW, particularly when composed of crushed bricks and old mortar, exhibits pozzolanic activity and filler effects that contribute to improved performance in geopolymer systems. Research has shown that CDW can enhance freeze-thaw resistance, reduce chloride permeability, and limit mass loss in sulphate-rich environments. However, many of these studies focus only on single exposure conditions or limited mechanical tests, without examining combined or long-term environmental effects that better represent field conditions summarised in Table 1 . Recent developments in machine learning (ML) have opened new opportunities in concrete research, especially in modelling compressive strength and mix optimisation [ 12 ]. Yet, the application of ML to predict durability parameters, such as strength retention after aggressive exposures, remains limited. Only a few studies have used SHAP analysis to clearly explain how different mix ingredients, such as the amount of Construction and Demolition Waste (CDW) or the activator-to-binder ratio, affect the performance of geopolymer concrete. This points to a noticeable gap in current research [ 12 ], [ 13 ]. Most existing studies focus either on experimental testing or on machine learning models separately, with few combining both approaches under different durability conditions. This study addresses that limitation by evaluating how geopolymer concrete containing CDW and artificial aggregates performs when exposed to aggressive environments. Machine learning is used not only to predict how the material behaves, but also to help understand which mix design factors have the greatest effect on strength retention [ 5 ]. Table 1 Comparative Review of Studies on Geopolymer Concrete (GPC) Durability and Machine Learning Applications References Materials Used Durability Parameters Studied Machine Learning Application Novelty Identified Research Gap [ 14 ] Glass powder, OPC Sulphate and chloride resistance None Comprehensive review of glass powder as a pozzolan No GPC integration or recycled aggregate study [ 15 ] Fly ash-based GPC Carbonation, sulphate, acid None Durability under chemical exposure Excludes recycled materials and ML integration [ 16 ] CDW aggregates, OPC Freeze–thaw, permeability None Durability of CDW concrete Not based on a geopolymer matrix [ 17 ] Recycled aggregates Shrinkage, strength, and workability None Partial replacement in OPC No ML or durability under aggressive exposures [ 18 ] Silica fume, fly ash Chloride ingress, compressive strength None Durability with SCMs No use of CDW or predictive modelling [ 19 ] Fly ash GPC Sulphate and acid resistance None GPC under aggressive media Excludes recycled aggregates [ 20 ] Fly ash, CDW Compressive strength, abrasion None CDW-based GPC No chemical/environmental durability study [ 21 ] Fly ash GPC Compressive strength ANN, SVR Applied ML for strength No durability or explainable AI tools [ 22 ] OPC concrete Compressive strength RF, ANN ML prediction model No focus on GPC or recycled materials [ 23 ] Glass powder, metakaolin Sulphate and acid None Blending recycled glass with binders No ML or CDW incorporation [ 24 ] Cementitious composites Strength and durability SHAP, LIME Explainable ML use Not focused on GPC or recycled aggregates [ 25 ] AAS + CDW Carbonation, RCPT None Alkali-activated concrete with CDW No ML application [ 26 ] Fly ash GPC Long-term sulphate resistance None Long-term study No AI or CDW inclusion [ 27 ] CDW, fly ash, GGBS Mechanical & durability None Comprehensive CDW–GPC assessment Lacks AI-based prediction tools This Study Fly ash, glass powder, CDW, artificial aggregates Sulphate, chloride, freeze–thaw, strength retention XGBoost, SHAP, ANN, SVR Integrates recycled CDW and explainable ML for harsh exposure Bridges experimental data with interpretable AI for sustainable GPC design 3. Materials and Methodology In this study, both lab testing and machine learning techniques were used to examine how durable geopolymer concrete (GPC) is when made with recycled construction and demolition waste (CDW) and artificial coarse aggregates. The testing process was designed to understand the specific role of recycled materials and to reflect how the concrete might perform over time in harsh environmental conditions. 3.1 Materials The binder used in the geopolymer mix was made by combining low-calcium Class F fly ash with ground waste glass. Both materials contain aluminosilicates, which react well with alkaline solutions and help form the structure of the hardened concrete. They were chosen because of their ability to improve strength and contribute to the chemical stability of the mix. CDW was collected from demolition sites and comprised primarily of crushed brick, mortar residues, and aged concrete. After screening and crushing, CDW was graded to fine aggregate size (passing 4.75 mm) and used as a partial or full replacement for natural sand. Artificial coarse aggregates, manufactured through sintering lightweight materials, were used across all mixes to ensure consistency in bulk density and enhance internal curing and thermal insulation properties [ 28 ], [ 29 ]. An alkaline activator solution, prepared with sodium hydroxide (10M concentration) and sodium silicate (approximately 2.0), served as the chemical trigger for the polymerisation process. The activator-to-binder (A/B) ratio was maintained at 0.45 throughout the study to ensure comparability across mixes. The oxide composition of both fly ash and glass powder, dominated by SiO₂, Al₂O₃, and minor CaO, was verified using X-ray fluorescence (XRF). Their reactivity contributes significantly to the formation of N-A-S-H gels, imparting mechanical and chemical stability under environmental stressors. The chemical composition of the materials used is as shown in Fig. 1 & Fig. 2 . The research methodology is as shown in Fig. 3 . 3.2 Mix Design Three distinct concrete mixtures were formulated to investigate the effect of CDW replacement levels, as shown in Table 2 . The proportions of fly ash and glass powder were held constant in all mixes to ensure that observed differences in performance could be attributed solely to CDW content. The mixing procedure involved dry blending of binders and aggregates, followed by gradual addition of the activator solution and thorough mechanical mixing to achieve homogeneity. Table 2 Different Mix Samples Mix ID CDW Content (% of Fine Aggregate) Fly Ash (%) Glass Powder (%) Activator/Binder Ratio Artificial Coarse Aggregate (%) GPC-0 0 70 30 0.45 100 GPC-50 50 70 30 0.45 100 GPC-100 100 70 30 0.45 100 3.3 Sample Preparation and Curing Cylindrical and prismatic specimens were cast using standard moulds (100 mm × 200 mm for cylinders and 100 mm × 100 mm × 500 mm for beams). All samples were demoulded after 24 hours and cured under ambient laboratory conditions (23 ± 2°C, 65% RH) for 28 days to simulate in-situ curing scenarios without artificial heat treatment. 3.4 Durability Testing Procedures To evaluate long-term performance, specimens were subjected to the following durability assessments. Immersion in a 5% Na₂SO₄ solution was conducted for up to 90 days, with periodic mass loss measurements at 28, 56, and 90 days. Samples underwent 300 cycles of freezing (-18°C) and thawing (+ 4°C) as per ASTM C666. Surface degradation was quantified by measuring surface scaling thickness. The Rapid Chloride Permeability Test (RCPT) was conducted under ASTM C1202, where the total charge passed (in Coulombs) was used to classify permeability levels. Post-exposure compressive strength tests were carried out to calculate strength retention ratios, offering a composite indicator of chemical and mechanical integrity [ 30 ]. 3.5 Machine Learning Modelling A structured dataset incorporated key variables like CDW percentage, A/B ratio, exposure type, and measured durability outcomes. Five regression algorithms were deployed for strength retention prediction are Linear Regression (LR), Support Vector Regression (SVR), Random Forest (RF), XGBoost, and Artificial Neural Networks (ANN). To measure how well the models performed, their results were compared using standard error metrics: R², Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). Among the models tested, XGBoost gave the most accurate predictions. This is likely because of how it builds trees step by step to reduce errors and avoid overfitting. The machine learning models were used to predict how the concrete would perform under tough environmental conditions. The focus was on estimating compressive strength, weight loss, and how much strength remained after exposure, using information about the mix design and type of exposure. 3.5.1 Data Preprocessing Before training the machine learning models, the experimental data were checked and organised. The inputs included details like how much fly ash and glass powder were used, the concentration of sodium hydroxide, the ratio of sodium silicate to sodium hydroxide, the amount of artificial and recycled aggregates, curing temperature, and how long the samples were exposed. The outputs being predicted were compressive strength (in MPa), percentage of weight loss, and strength remaining after exposure. Any errors or missing values in the data were removed to ensure consistency. Using a common scaling technique, all input values were made to fall within the same range, from 0 to 1, to facilitate comparison of the variables. To prevent bias during model training, this was required because variables with wider numeric ranges might The data used in this study came from a comprehensive experimental program that included 90 distinct samples produced by combining three mix designs (GPC-0, GPC-50, and GPC-100) and several exposure times under freeze-thaw, sulphate, and chloride conditions. Input features like CDW content, fly ash percentage, glass powder percentage, activator-to-binder (A/B) ratio, curing regime, and exposure type were included in every sample. Compressive strength, residual strength, and mass loss following environmental exposure were the output variables. Based on previous research and SHAP analysis, these inputs were chosen because of their known impact on durability. 3.5.1.1 Data Cleaning and Scaling All input features were normalised to a 0–1 range using Min-Max scaling before the model was trained. By taking this step, the learning process is prevented from being dominated by features with higher numerical values, like compressive strength or exposure duration. Because of the controlled experimental conditions, there were no missing or unusual data points. To ensure compatibility with all regression models, one-hot encoding was used to encode categorical variables, such as exposure type. 3.5.2 Model Selection and Hyperparameter Tuning Gradient Boosting Regressor (GBR), Random Forest Regressor (RFR), and Decision Tree Regressor (DTR) were the models utilised in this investigation. These models were chosen because of their resilience to outliers and capacity to manage non-linear relationships. Optimised hyperparameters were used during training for each model to increase accuracy and avoid overfitting. Grid search and five-fold cross-validation were used to adjust the hyperparameters. Table 3 displays the chosen hyperparameters. Using a grid search approach in conjunction with five-fold cross-validation, hyperparameter tuning was carried out to guarantee the best possible predictive accuracy across models. To determine which configuration produced the lowest error on validation sets, this method entailed methodically testing combinations of parameter values. Each model type had a different search range. No tuning was necessary for Linear Regression (LR). Support Vector Regression (SVR) was tuned for kernel type (linear, polynomial, RBF), regularisation parameter C (range: 0.1–100), and epsilon margin (0.01–1.0). Random Forest (RF) was tuned for the number of estimators (50–200), maximum depth (4–12), and minimum samples per split (2–5). For XGBoost, the learning rate (0.01–0.2), maximum depth (3–10), and number of estimators (50–200) were explored. Artificial Neural Network (ANN) tuning involved the number of hidden layers (1–3), neurons per layer (16–64), learning rate (0.001–0.01), and batch size (8–32). These parameters were selected based on prior studies in concrete prediction tasks and refined using experimental error analysis. Table 3 Optimised Hyperparameters for Each ML Model Model Key Hyperparameters Decision Tree Regressor Max depth = 6, Min samples split = 4 Random Forest Regressor Estimators = 100, Max depth = 8, Min samples split = 3 Gradient Boosting Regressor Learning rate = 0.1, Estimators = 150, Max depth = 5 Durability prediction model selection was performed in two stages to determine which machine learning model would be the best fit. Initially, in the first stage, five most popular regression models, specifically, Linear Regression, Support Vector Regression (SVR), Random Forest, XGBoost, and Artificial Neural Networks (ANN) were fitted to the standard settings. This comparative study at the first stage helped to find out which methods were consistently accurate with the lowest error. Then, out of all the rest models, Random Forest and XGBoost exhibited the highest degrees of protecting data from complexities concerning the error rate and were thus the most acceptable. At this further stage, three of the tree-based models: Decision Tree Regressor, Random Forest Regressor, and Gradient Boosting Regressor were thoroughly tuned in their parameters to identify the best performing model. The models were selected for their ability to capture non-linear patterns and for their increased model training control. Their parameters were tuned via grid search and cross-validation, and the final configurations are provided in Table 3 . Among those, XGBoost has the highest scores, so further investigation using SHAP will be conducted to make the input attribute effect clearer for the durability prediction model. 3.5.3 Cross-Validation Strategy To check how well the models could make predictions on new data, a five-fold cross-validation method was used. The dataset was randomly split into five parts. The model was trained in four parts per round, and it was tested in the fifth part. In order to test every component of the data once, this procedure was carried out five times. The average of the five rounds was used to determine the final outcomes. This method reduces the likelihood that the model will simply memorise the training set and helps demonstrate how the model performs on data it hasn't seen before. 3.6 SHAP-Based Feature Importance SHAP analysis was used to gain a better understanding of the factors that affected the machine learning predictions. This method demonstrated the relative contributions of each input variable to the model's output. It was discovered that the activator-to-binder ratio and the quantity of CDW had the biggest effects on the retention of compressive strength. The SHAP results helped to clarify how the XGBoost model arrived at its conclusions and were in good agreement with experimental trends. The biggest impact was from CDW content. As CDW replacement rose, strength retention generally improved, particularly up to roughly 70%. Following that, the benefit began to plateau or decline slightly, perhaps as a result of modifications to the internal structure or mix workability. The next most crucial element was the activator-to-binder ratio. A more compact and long-lasting concrete was produced when this ratio was balanced because it enhanced the chemical reaction between the binder and the activator. Another important factor was the kind of environmental exposure. Chloride was the exposure type that weakened the concrete the most, followed by sulphate and freeze-thaw. This is probably because chloride ions are highly reactive and can degrade the geopolymer matrix more forcefully than other agents. 4. Results and Discussion 4.1 Experimental Durability Performance The durability assessment of geopolymer concrete (GPC) incorporating varying proportions of construction and demolition waste (CDW) and artificial aggregates revealed clear performance trends under aggressive environmental exposures ( Fig. 4 ). As the CDW content increased, the mass loss after 90 days of sulphate exposure decreased notably, from 6.5% at 0% CDW to 3.7% at 100% CDW. This trend signifies enhanced sulphate resistance, likely due to the absence of Ca (OH)₂ in GPC and the pozzolanic contribution of crushed brick in CDW. Surface scaling under freeze-thaw cycles also diminished with higher CDW content, dropping from 2.7 mm to 1.4 mm. The enhanced freeze-thaw resistance can be attributed to the lower permeability and refined pore structure imparted by recycled fines and the lightweight artificial coarse aggregates. Strength retention showed a positive correlation with CDW content. Samples with 70–100% CDW retained up to 88% of their original compressive strength after sulphate attack, compared to just 72% in CDW-free mixes. Additionally, Rapid Chloride Penetration Test (RCPT) results indicated a reduction in chloride ion permeability, with charge passed decreasing from 1400 Coulombs (moderate permeability) at 0% CDW to 850 Coulombs (low permeability) at 100% CDW. 4.1.1 Mass Loss under Sulphate Attack After 90 days of exposure to a 5% Na ₂ SO₄ solution, mass loss exhibited a clear declining trend with increasing CDW content. Specifically, the control mixture with 0% CDW lost approximately 6.5% of its mass, whereas the mixture composed entirely of CDW lost only about 3.7%. The intermediate mixture with 50% CDW lost about 5.0% over the same period. For example, at 28 days, the mass losses were 2.8% (0% CDW), 2.2% (50% CDW), and 1.7% (100% CDW). At 56 days, the corresponding losses were 5.2%, 4.1%, and 3.0%, respectively. These results are shown in Table 4 , that incorporating CDW substantially mitigates the destructive effects of sulphate attack on the material. This improvement may be attributed to the consumption of portlandite (Ca (OH)₂) and the resultant densification of the microstructure, thereby reducing the material’s susceptibility to sulphate-induced deterioration. Table 4 Mass Loss of Geopolymer Concrete Under Sulphate Attack at Different CDW Replacement Levels Over Time Exposure Time (Days) Mass Loss 0% CDW (%) Mass Loss 50% CDW (%) Mass Loss 100% CDW (%) 28 2.8 2.2 1.7 56 5.2 4.1 3.0 90 6.5 5.0 3.7 4.1.2 Surface Scaling under Freeze-Thaw Cycles Following 300 cycles of freezing and thawing, the extent of surface scaling was observed to decrease with increasing CDW content. For instance, the control specimens with no CDW exhibited severe scaling of approximately 2.7 mm, whereas those composed entirely of CDW showed only minor scaling of around 1.4 mm. This enhancement in freeze-thaw resistance can be attributed to reduced capillary porosity and the densification of the microstructure due to pozzolanic reactions contributed by the recycled brick fines. Although brick fines are hydrophilic and naturally absorb more water, their fine particle size and reactive silica-alumina content can help fill voids and enhance gel formation within the matrix. This leads to reduced interconnected capillary pores, which limit free water movement and mitigate internal pressure during freezing cycles introduced by the recycled brick fines. These factors inhibit water ingress and ice formation within the pore structure, thereby mitigating surface damage under cyclic freeze-thaw conditions. Although both CDW and artificial lightweight aggregates are known to have higher intrinsic porosity and water absorption compared to natural aggregates, their influence on freeze–thaw resistance is more nuanced. Several studies [Silva et al., 2021; Li et al., 2018] have shown that when used in geopolymer systems, the enhanced microstructural bonding between alkali-activated binders and recycled brick particles can reduce capillary pore continuity, thereby limiting water ingress and internal pressure build-up during freeze–thaw cycles. Moreover, the internal curing effect of lightweight aggregates can help buffer moisture fluctuations by gradually releasing absorbed water, reducing pore saturation levels during freezing events [ 30 ], [ 31 ]. As a result, the apparent permeability may not fully reflect the effective transport pathways relevant to freeze–thaw deterioration. Thus, despite higher absorption values, the observed surface scaling reduction can be attributed to improved pore structure refinement and internal moisture regulation mechanisms, consistent with findings from recent geopolymer durability research [ 32 ], [ 33 ]. 4.1.3 Strength Retention The retention of compressive strength after sulphate exposure was found to be highly dependent on the level of CDW replacement. Geopolymer concrete mixtures containing 70–100% CDW retained approximately 85–88% of their original compressive strength, whereas the CDW-free mixture retained only about 72%. These observations imply that higher CDW content leads to a geopolymer matrix that is chemically more stable and mechanically more resilient under aggressive sulphate attack shown in Fig. 5 . The findings indicate that the inclusion of recycled construction waste markedly bolsters the material’s resistance to strength loss in chemically aggressive environments. 4.1.4 Chloride Ion Penetrability (RCPT) The rapid chloride permeability test (RCPT) results indicate that chloride ion permeability decreased markedly as CDW content increased. In practical terms, the 0% CDW mixture allowed a charge of about 1400 Coulombs to pass (classified as moderate permeability by ASTM C1202), whereas the 100% CDW mixture allowed only 850 Coulombs (classified as low permeability). This significant reduction in permeability is attributed to the formation of a denser geopolymer matrix, resulting from the pozzolanic reactivity and filler effects of the recycled brick materials. Incorporating CDW into the geopolymer matrix substantially hinders the penetration of chloride ions by creating a more compact pore structure. 4.2 Machine Learning Model Prediction To assess the predictive reliability of different machine learning (ML) models for estimating compressive strength retention in geopolymer concrete, five algorithms were trained and validated using experimental data. The models included Linear Regression, Random Forest, Support Vector Regression (SVR), XGBoost, and Artificial Neural Networks (ANN). Their performances were compared using three widely recognised statistical metrics: The coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE), as shown in Fig. 6 & Table 5 . The tested models, XGBoost, showed the highest predictive accuracy, achieving an R² value of 0.96, coupled with a low MAE of 1.1 and RMSE of 1.6. These results indicate its exceptional ability to learn and generalise from complex patterns present in the input variables, including CDW content, activator ratios, and environmental exposure types. The Random Forest and ANN models also showed good results, with R² of 0.92 and 0.94. The models are also able to handle non-linear relationships. The Random Forest and ANN models are less accurate than the XGBoost model. The Random Forest model is also characterized by a high level of interpretability and robustness to overfitting. The ANN model is rather flexible and may suffer from variations in the training process. The Support Vector Regression model has R² <0.88, and its performance is less effective compared to other models. In particular, it was difficult to set the hyperparameters in the case of this model. Thus, the model is sensitive to the features of the dataset. The Linear Regression model demonstrated the lowest level of accuracy. Although this model also has R² of 0.78, it is not able to find dependencies in the case of non-linear data. This model cannot be used for this dataset. The MAE and RMSE values are 3.2 and 4.8, respectively. To assess the predictive accuracy of the machine learning models, standard statistical metrics were employed. The coefficient of determination (R²) measures how well the model explains variance in the observed data. MAE quantifies the average magnitude of errors, while RMSE penalizes larger deviations more heavily. Coefficient of Determination (R²) = 1 - Σ(yi - ŷi)² / Σ(yi - ȳ)² Eq. (1) Mean Absolute Error (MAE) = (1/n) Σ|yi - ŷi| Eq. (2) Root Mean Square Error (RMSE) = sqrt((1/n) Σ(yi - ŷi)²) Eq. (3) Table 5 Model Performance Comparison Model R² (Strength) MAE RMSE Linear Regression 0.78 3.2 4.8 Random Forest 0.92 1.4 2.1 Support Vector Regression (SVR) 0.88 1.9 2.6 XGBoost 0.96 1.1 1.6 Artificial Neural Network (ANN) 0.94 1.3 1.9 To improve the predictive accuracy, each machine learning model's hyperparameters were adjusted. Because of their capacity to handle intricate non-linear patterns in data while simultaneously lowering overfitting rates, the Random Forest and XGBoost models distinguished themselves with their robust and consistent performance. To increase the accuracy of the XGBoost model, important parameters like learning rate, tree depth, and the number of estimators were changed. Even when applied to previously untested data, these modifications improved the model's ability to predict compressive strength. The model's stability was examined using cross-validation, which made sure it functioned well in various dataset sections. Even though XGBoost produced the most accurate results, some people find it challenging to understand. To solve this, the model's prediction process was examined using SHAP analysis (described in Section 4.3 ). The activator-to-binder ratio, the type of environmental exposure, and the percentage of construction and demolition waste (CDW) were among the most crucial input variables that were identified with the aid of this technique. These results improved mix design choices in geopolymer concrete and helped explain the model's decisions. Future Directions: In the future, it may be useful to explore hybrid models that combine different machine learning techniques to improve predictive accuracy. For instance, stacking or assembling XGBoost with Random Forest could provide a more robust prediction framework. Furthermore, integrating explainability techniques such as SHAP with machine learning models will continue to support more transparent and data-driven decisions in geopolymer concrete mix design. 4.2.1 Comparison Between Experimental Results and ML Predictions (GBR Model) To further verify the performance of the Gradient Boosting model, predicted outputs were compared with experimental results for compressive strength (CS), mass loss (ML), and residual strength (RS) under harsh exposure conditions. The results are presented in Table 6 , where mix proportions follow the same categorisation as described in Table 2 . The machine learning predictions showed excellent agreement with the experimental results across all mix types (GPC-0, GPC-50, and GPC-100). The slight deviations were within acceptable margins, confirming the robustness of the model. Furthermore, the effectiveness of the model is visually reinforced in Figs. 7 , 8 & Fig. 9 , where predictions closely follow the experimental trend line Table 6 Experimental vs ML Prediction Results for Different Mix Proportions Mix ID CDW (%) Fly Ash (%) Glass Powder (%) Activator/Binder Ratio Artificial Aggregate (%) CS Exp (MPa) CS Pred (MPa) ML Exp (%) ML Pred (%) RS Exp (MPa) RS Pred (MPa) GPC-0 0 70 30 0.45 100 42.5 41.9 5.2 5.5 35.6 34.8 GPC-50 50 70 30 0.45 100 38.4 39 6.1 6 32 31.5 GPC-100 100 70 30 0.45 100 46.2 45.7 4.3 4.6 37.4 36.9 Table 6 presents a comparative analysis of experimental and machine learning (ML)-predicted results for three different geopolymer concrete (GPC) mix designs incorporating varying percentages of construction and demolition waste (CDW). The mix IDs—GPC-0, GPC-50, and GPC-100—denote 0%, 50%, and 100% replacement of natural fine aggregates with CDW, respectively, while maintaining consistent proportions of fly ash (70%) and glass powder (30%) as binders, an activator-to-binder ratio of 0.45, and a fixed artificial aggregate content of 100%. The compressive strength (CS) values predicted by the machine learning model were very close to the results from the lab tests. For example, in the GPC-0 mix, the measured strength was 42.5 MPa, and the model gave a value of 41.9 MPa. Similar agreement was seen in GPC-50 and GPC-100, where the predicted strengths were 39.0 MPa and 45.7 MPa, compared to experimental values of 38.4 MPa and 46.2 MPa. This shows that the model worked well across different levels of CDW replacement and was able to make accurate predictions for various mix designs. The model demonstrated good performance for mass loss (ML) under aggressive conditions. The largest discrepancy between expected and actual values was 0.3%, indicating that the model did a good job of capturing the effects of environmental exposure. For example, the model predicted an ML of 4.6% in the GPC-100 mix, while the measured ML was 4.3%. Given that safety margins are frequently preferred in real-world scenarios, this slight overestimation might be helpful. The same pattern was also seen in the residual strength (RS) predictions following exposure. The experimental value of RS for GPC-0 was 35.6 MPa, a mere 2.3% difference from the predicted value of 34.8 MPa. For other mixes, similar outcomes were obtained, indicating that the model can accurately predict strength and durability performance following exposure. Across various geopolymer concrete mixes, the machine learning model performed consistently. The model can be used to estimate important properties of concretes made with recycled and artificial materials, according to the slight variations between the expected and actual results. When designing mixes that must satisfy strength and durability requirements in sustainable construction, this type of prediction is helpful. When predicting the durability of geopolymer concrete containing CDW and artificial aggregates, XGBoost produced the most accurate results out of all the models that were tested. The operation of XGBoost is responsible for this high accuracy. It gradually improves predictions, learning from past errors to make better choices in subsequent iterations. Additionally, it has built-in controls to prevent overfitting by preventing the model from becoming overly complex. XGBoost can automatically determine which input features are most important, handle missing values, and handle both simple and complex input patterns. Because of these characteristics, it works particularly well when working with concrete data, where there are frequently intricate and non-linear relationships between ingredients and performance. 4.3 SHAP-Based Feature Importance Analysis SHAP analysis was used to improve the XGBoost model's predictions' interpretability. This method revealed the underlying relationships between mix design parameters and durability outcomes by quantifying the individual contributions of each input variable to the prediction of compressive strength retention. The most important factor influencing the model's output among the input features was the CDW content. It exerted a strong positive influence on durability, particularly up to a replacement level of approximately 70%, beyond which the marginal gains diminished. This plateau effect suggests that while CDW can enhance the packing density and contribute to matrix densification at moderate levels, excessive substitution may introduce variability or adverse effects on matrix integrity. The activator-to-binder ratio was identified as the second most influential parameter. Its prominence underscores its role in governing the dissolution rate of aluminosilicates and subsequent geo-polymerisation. Optimal ratios were found to promote better gel formation and pore structure refinement, thereby enhancing resistance to environmental degradation. The type of exposure condition, whether chloride, sulphate, or freeze–thaw, also significantly influenced strength retention. Among these, chloride attack was the most detrimental, followed by sulphate and freeze–thaw exposure. The SHAP-based insights corroborate the experimental observations and reinforce the importance of balanced mix design. Specifically, they highlight the need to optimise CDW usage and alkali activator chemistry to develop durable, low-carbon concretes capable of withstanding harsh service environments. This type of feature-level explainability is critical for promoting the use of smart, data-driven mix design strategies in sustainable infrastructure development. The bar chart titled "SHAP-Based Feature Importance in XGBoost Model" illustrated in Fig. 10 , shows that the different input variables contribute to predicting the compressive strength retention of geopolymer concrete under harsh environmental conditions. Each bar shows the mean absolute SHAP value for a feature, indicating its average effect on the model’s output across the dataset. Among all the input features, the content of Construction and Demolition Waste (CDW) had the highest SHAP value (0.35), showing that it had the strongest effect on durability, particularly when used in suitable amounts. With a SHAP value of 0.25, the Activator-to-Binder Ratio followed, as it is crucial to the way the chemical reactions occur during the formation of geopolymers. Third place went to Exposure Type (0.20), which illustrates how various environmental factors, like sulphate, chloride, or freeze-thaw cycles, can impact the performance of concrete. The curing regime contributed to the early strength development and long-term stability of the geopolymer mix despite having a lower SHAP value (0.10). Fly ash (0.04) and glass powder (0.06) had less of an impact but were still involved in the concrete's overall behavior. This kind of analysis can be helpful in enhancing mix designs by highlighting the mix components that have the biggest effects on strength retention. These findings also lend credence to sensible choices that strike a balance between sustainability and durability in practical applications. According to the SHAP analysis, one of the key variables influencing durability predictions was the sodium-to-aluminum (Na/Al) molar ratio. This outcome is in good agreement with current knowledge of geopolymer chemistry. To create a dense aluminosilicate gel that increases strength and resistance to chemical attack, a balanced Na/Al ratio is required. Excessive sodium levels can cause surface problems like efflorescence and leave behind unreacted alkalis. On the other hand, too little sodium may result in incomplete polymerization and a weaker internal structure. The type and amount of artificial aggregate also ranked highly, likely due to their effect on how pores form and how well the aggregate bonds with the binder. A SHAP dependence plot for the Na/Al ratio showed that its influence was not linear. The plot revealed a peak effect on durability when the ratio was between about 0.9 and 1.2. Outside this range, the predicted benefit dropped off. This trend reflects how important it is to maintain the right balance of alkalis. If the sodium content is too high, the gel may become unstable, allowing ions to move more freely and possibly leading to cracks from repeated wetting and drying. If it’s too low, the geopolymerization reaction may not fully develop, resulting in a porous and weaker material. This type of model-based insight helps link prediction outcomes to actual material behavior and can support better decision-making in concrete design. The SHAP dependence plot for artificial fine aggregate replacement showed that durability improved as the replacement level increased, especially up to around 50%. After that point, the improvement slowed or slightly decreased. This pattern suggests that replacing natural sand with artificial fine aggregates from construction and demolition waste can help improve the internal structure of the concrete. A moderate amount of replacement appears to fill voids more effectively and reduce the number of tiny pores that allow water and chemicals to move through the material, making it more resistant to environmental damage. The presence of angular particles with rough surfaces likely improves the interfacial transition zone (ITZ) between the binder and aggregate phases, contributing to superior mechanical interlock and reduced permeability. However, excessive replacement may introduce inconsistencies in particle size distribution or residual contaminants, which could compromise homogeneity and lead to localized weakness. Thus, the SHAP-based interpretation highlights the importance of optimizing the aggregate replacement ratio to balance sustainability goals with long-term performance. 4.3.1 Environmental Exposure Condition, Specifically the Number of Wet-Dry Cycles, As Interpreted from the SHAP Dependence Plot The SHAP dependence analysis for the number of wet-dry cycles revealed a consistently negative impact on the predicted durability index, with SHAP values decreasing steadily as the number of cycles increased. This outcome is consistent with known deterioration mechanisms in geopolymer concrete subjected to cyclic moisture fluctuations. Repeated wetting and drying accelerate the ingress of water and dissolved ions, which can cause internal microcracking, leaching of alkali ions, and progressive weakening of the aluminosilicate gel matrix. The SHAP dependence plot thus reflects the physical degradation pathways that manifest under prolonged environmental stress. Notably, the rate of SHAP value decline appeared to accelerate beyond 20 cycles, suggesting a threshold after which damage mechanisms become more aggressive or irreversible. This insight underscores the model’s ability to capture not just statistical correlations but meaningful durability trends under real-world exposure, thereby enhancing its practical utility in performance-based mix design. 4.3.2 Dependence of Model Predictions on Durability Figure 11 illustrates the SHAP dependence plot for Durability, which emerged as the most influential feature in the XGBoost model. The plot demonstrates a clear positive correlation between the standardized durability values and the corresponding SHAP values. This indicates that as durability increases, the model's prediction for fatigue life also increases, signifying a strong and consistent contribution of durability to the output variable. The scatter pattern reveals a relatively linear relationship, suggesting minimal interaction effects within the observed range, and affirming that durability serves as a stable and independent predictor in the trained model. In line with previous test results and engineering practice, the model indicates that longer service life is associated with higher durability values when the material is subjected to repeated loading. The distribution of the data points further demonstrates the model's adaptability to various scenarios. This demonstrates that the model is capturing relationships that make sense from a physical and engineering point of view and supports the notion that durability has a significant impact on how materials behave under fatigue. The comparison of different models is shown in Table 7 . Table 7 Comparison of Machine Learning Models for Predicting Concrete Durability Study / Reference Concrete Type Input Features ML Model Used Model Interpretability Reported Prediction Accuracy [ 34 ] Fly ash-based geopolymer concrete Mix proportions, curing conditions Linear Regression Low Moderate [ 35 ] OPC concrete under sulphate exposure Water–cement ratio, aggregate properties Random Forest Moderate High [ 36 ] Recycled aggregate concrete Porosity, slump, compressive strength Support Vector Machine Low Moderate This Study (2025) Geopolymer concrete with CDW & AA Water absorption, capillary uptake, acid resistance, mix ratios XGBoost + SHAP High High 5. Conclusion This study confirms that machine learning models can reliably predict the durability of geopolymer concrete made with CDW and artificial aggregates. Among the five models tested, XGBoost delivered the most accurate results, with an R² value of 0.96, a mean absolute error (MAE) of 1.1, and a root mean square error (RMSE) of 1.6. These results closely matched the experimental data, such as compressive strength values of 42.5 MPa (experiment) vs. 41.9 MPa (prediction) for GPC-0, and 46.2 MPa vs. 45.7 MPa for GPC-100. The durability results also showed a clear benefit from using CDW. Strength retention after sulphate exposure reached up to 88% for high-CDW mixes, while chloride ion permeability dropped from 1400 to 850 Coulombs as CDW content increased. SHAP analysis helped explain the role of different variables, with CDW content and the activator-to-binder ratio having the greatest influence on durability. Overall, combining experimental testing with interpretable ML tools offers a practical way to design more sustainable and long-lasting concrete mixes, particularly for structures exposed to aggressive conditions like sulphate, chloride, or freeze–thaw environments. 6. Future Scope Building upon the outcomes of this study, several directions are recommended for future research. First, while the current work focused on the durability of geopolymer concrete under sulphate, freeze–thaw, and chloride exposures, future studies may consider additional degradation mechanisms such as carbonation, alkali–silica reaction (ASR), and combined environmental attacks to assess long-term performance comprehensively. Furthermore, life cycle assessment (LCA) and cost-benefit analysis of CDW-based GPC mixtures would offer critical insights into their environmental and economic viability on a broader scale. From a modelling perspective, integrating ensemble or hybrid machine learning approaches, such as model stacking or deep learning frameworks, may enhance predictive accuracy and generalisation across varied datasets. Additionally, incorporating real-time monitoring data from field applications could bridge the gap between laboratory-scale experimentation and practical implementation. Exploring explainable AI (XAI) tools beyond SHAP, such as LIME or counterfactual analysis, may also provide deeper interpretability of model predictions for decision-making in mix design. Lastly, scaling up the application of CDW-incorporated GPC in structural elements, pavements, or marine infrastructure, coupled with rigorous structural performance testing, will be critical for validating the suitability of this sustainable material in diverse civil engineering contexts. Such efforts will support the transition toward resilient, low-carbon construction practices aligned with global sustainability goals. Although the model achieved high accuracy, the dataset was limited to a specific set of materials and conditions. Future research could explore larger, multi-source datasets and additional environmental exposure types to enhance generalizability Abbreviations GPC: Geopolymer Concrete, CDW: Construction and Demolition Waste, ML: Machine Learning, XGB / XGBoost: eXtreme Gradient Boosting, RF: Random Forest, ANN: Artificial Neural Network, SVR: Support Vector Regression, LR: Linear Regression, R²: Coefficient of Determination, RMSE – Root Mean Square Error, SHAP – SHapley Additive exPlanations, NaOH: Sodium Hydroxide, Na 2 SiO 3 : Sodium Silicate, A/B : Activator-to-Binder ratio, Declarations Conflicts of Interest: No Conflicts of Interest. Author Contribution Atul S. Kurzekar: Conceptualization, Methodology, Writing Original Draft, Software. Uday P. Waghe: Methodology, Investigation, Writing, Review & Editing. Prajakta Waghe: Formal Analysis, Validation, Resources. Data Availability: The data sets used and/or analysed during the current study are available from the Corresponding author on reasonable request. References Mohajerani A et al (2019) Recycling waste materials in geopolymer concrete. 10.1007/s10098-018-01660-2 Gunasekara C, Setunge S, Law DW, Willis N, Burt T (2018) Engineering Properties of Geopolymer Aggregate Concrete. J Mater Civ Eng 30(11). 10.1061/(asce)mt.1943-5533.0002501 Christensen TH, Birgisdottir H (2010) Recycling of Construction and Demolition Waste, in Solid Waste Technology & Management , vol. 1. 10.1002/9780470666883.ch19 Guo H et al (2018) Durability of recycled aggregate concrete – A review. Cem Concr Compos 89. 10.1016/j.cemconcomp.2018.03.008 Janiesch C, Zschech P, Heinrich K (2021) Machine learning and deep learning. Electron Markets 31(3). 10.1007/s12525-021-00475-2 Sarker IH (2021) Machine Learning: Algorithms, Real-World Applications and Research Directions. 10.1007/s42979-021-00592-x Alomari Y, Andó M (2024) SHAP-based insights for aerospace PHM: Temporal feature importance, dependencies, robustness, and interaction analysis. Results Eng 21. 10.1016/j.rineng.2024.101834 8, Atul S, Kurzekar S, Kurzekar et al (2020) An Investigation on Structural Performance of Tall Building Embedded with Core, Outrigger System. Int J Mech Prod Eng Res Dev 10(3):2043–2048. 10.24247/ijmperdjun2020191 Shivaprasad KN, Das BB (2018) Determination of optimized geopolymerization factors on the properties of pelletized fly ash aggregates. Constr Build Mater 163. 10.1016/j.conbuildmat.2017.12.038 Gerges NN, Issa CA, Fawaz SA, Jabbour J, Jreige J, Yacoub A (2018) Recycled Glass Concrete: Coarse and Fine Aggregates. Eur J Eng Technol Res 3(1). 10.24018/ejeng.2018.3.1.533 Patel S, Pansuria A, Shah V, Patel S (2014) Construction and Demolition Waste Recycling. IJIRST –International J Innovative Res Sci Technology|, 1, 7 Mahesh B (2020) Machine Learning Algorithms - A Review. Int J Sci Res (IJSR) 9(1). 10.21275/art20203995 Paleyes A, Urma RG, Lawrence ND (2022) Challenges in Deploying Machine Learning: A Survey of Case Studies. ACM Comput Surv 55(6). 10.1145/3533378 Rashidian-Dezfouli H, Rao Rangaraju P (2018) Evaluation of Selected Durability Properties of Portland Cement Concretes Containing Ground Glass Fiber as a Pozzolan, Transportation Research Record: Journal of the Transportation Research Board , vol. 2672, no. 27, pp. 88–98, Dec. 10.1177/0361198118773198 Bernal SA, Rodríguez ED, Mejía R, de Gutiérrez, Provis JL (2012) Performance of alkali-activated slag mortars exposed to acids, J Sustain Cem Based Mater , vol. 1, no. 3, pp. 138–151, Sep. 10.1080/21650373.2012.747235 Patil AS, Kurzekar AS, Fulzele P (2019) Behaviour of H sections with slender flange built upsection in pre engineering building. J Adv Res Dyn Control Syst, 11, no. 8 Special Issue Poon C-S, Chan D (May 2007) The use of recycled aggregate in concrete in Hong Kong. Resour Conserv Recycl 50(3):293–305. 10.1016/j.resconrec.2006.06.005 Siddique R, Khatib J, Kaur I (2008) Use of recycled plastic in concrete: A review. Waste Manag 28(10):1835–1852. 10.1016/j.wasman.2007.09.011 Chokkalingam RB, Ganesan N (2017) A study on the strength development of geopolymer concrete using fly ashing fly ash. Int J Eng Technology(UAE) 6(4). 10.14419/ijet.v6i4.7761 Sherwani AFH, Faraj R, Younis KH, Daraei A (2021) Strength, abrasion resistance and permeability of artificial fly-ash aggregate pervious concrete. 10.1016/j.cscm.2021.e00502 Nguyen-Sy T, Wakim J, To Q-D, Vu M-N, Nguyen T-D, Nguyen T-T (Nov. 2020) Predicting the compressive strength of concrete from its compositions and age using the extreme gradient boosting method. Constr Build Mater 260:119757. 10.1016/j.conbuildmat.2020.119757 Waghmare C et al (2025) Machine learning based prediction of compressive strength in roller compacted concrete: a comparative study with PDP analysis. Asian J Civil Eng. 10.1007/s42107-025-01310-1 Singh H, Siddique R (Oct. 2023) Long term durability assessment of self-compacting concrete made with crushed recycled glass and metakaolin. Constr Build Mater 400:132656. 10.1016/j.conbuildmat.2023.132656 Okasha NM, Mirrashid M, Naderpour H, Ciftcioglu AO, Meddage DPP, Ezami N (2024) Machine learning approach to predict the mechanical properties of cementitious materials containing carbon nanotubes, Developments in the Built Environment , vol. 19, p. 100494, Oct. 10.1016/j.dibe.2024.100494 Yaphary YL, Lau D, Sanchez F, Poon CS (May 2020) Effects of sodium/calcium cation exchange on the mechanical properties of calcium silicate hydrate (C-S-H). Constr Build Mater 243:118283. 10.1016/j.conbuildmat.2020.118283 Bilir T, Gencel O, Topcu IB (2015) Properties of mortars with fly ash as fine aggregate. Constr Build Mater 93. 10.1016/j.conbuildmat.2015.05.095 Naveena MP, Narayana G, Ravichandra R (2017) Study on strength and durability properties of fly ash aggregates concrete. Indian Concrete J, 91, 7 Kurzekar AS, Waghe U, Gawande D, Jaiswal A, Narkhede P (2025) Examining Fine Aggregate Properties from Dismantled Building Materials Waste: Towards Sustainable Construction Solutions in Environmental Sensing. 1–8. 10.1007/978-3-031-73816-6_1 Kurzekar AS, Waghe UP, Bhagat RM, Kherde YP (2024) Testing of mortar properties using dismantled building material and deteriorated bricks as substitution of natural fine aggregate for the reduction of cement. 070005. 10.1063/5.0240286 Jain KL, Sancheti G, Gupta LK (2020) Durability performance of waste granite and glass powder added concrete. Constr Build Mater 252. 10.1016/j.conbuildmat.2020.119075 Saravanakumar P (2018) Strength and durability studies on geopolymer recycled aggregate concrete. Int J Eng Technology(UAE) 7(2). 10.14419/ijet.v7i2.24.12087 Masood B, Elahi A, Barbhuiya S, Ali B (2020) Mechanical and durability performance of recycled aggregate concrete incorporating low calcium bentonite. Constr Build Mater 237. 10.1016/j.conbuildmat.2019.117760 Shaikh FUA (2016) Mechanical and durability properties of fly ash geopolymer concrete containing recycled coarse aggregates. Int J Sustainable Built Environ 5(2). 10.1016/j.ijsbe.2016.05.009 Ahmad A, Ahmad W, Aslam F, Joyklad P (Jun. 2022) Compressive strength prediction of fly ash-based geopolymer concrete via advanced machine learning techniques. Case Stud Constr Mater 16:e00840. 10.1016/j.cscm.2021.e00840 Elahi MMA et al (Apr. 2021) Improving the sulfate attack resistance of concrete by using supplementary cementitious materials (SCMs): A review. Constr Build Mater 281:122628. 10.1016/j.conbuildmat.2021.122628 Peng Y, Unluer C (Mar. 2023) Modeling the mechanical properties of recycled aggregate concrete using hybrid machine learning algorithms. Resour Conserv Recycl 190:106812. 10.1016/j.resconrec.2022.106812 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 04 Aug, 2025 Read the published version in Iranian Journal of Science and Technology, Transactions of Civil Engineering → Version 1 posted Editorial decision: Revision requested 11 Jul, 2025 Reviews received at journal 03 Jul, 2025 Reviewers agreed at journal 18 Jun, 2025 Reviews received at journal 17 Jun, 2025 Reviews received at journal 16 Jun, 2025 Reviewers agreed at journal 16 Jun, 2025 Reviewers agreed at journal 16 Jun, 2025 Reviewers agreed at journal 16 Jun, 2025 Reviewers invited by journal 16 Jun, 2025 Editor assigned by journal 10 Jun, 2025 Submission checks completed at journal 10 Jun, 2025 First submitted to journal 10 Jun, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6862865","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":472222600,"identity":"59f29f40-0723-4bf7-9e39-7993c5ca707c","order_by":0,"name":"Atul S. Kurzekar","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4UlEQVRIiWNgGAWjYFACNjBpwMDAfAAqkkC0FjaYUuK18BgQp4W//1iadMUfG2P+2We+SfyosWHgZ88xwKtF4kbaMckzPGlmEudyt0n2HEtjkOx5g18Lww32NskGicM2DGd4t0kzNhxmMLhBwBb588eBWgz+28if4XkG1PKfwZ6QFoMDQIc1JBwwMzjDwwbUcoDBQIKAFsMbacmWDQeSjQ3PsBlb9hxL5pE486wArxa588cMbzb8sTOcd4b54Y0fNXZy/O3JG/BqQQYsEkCCh2jlIMD8gSTlo2AUjIJRMGIAAII0QzEFoMhsAAAAAElFTkSuQmCC","orcid":"","institution":"Yeshwantrao Chavan College of Engineering","correspondingAuthor":true,"prefix":"","firstName":"Atul","middleName":"S.","lastName":"Kurzekar","suffix":""},{"id":472222601,"identity":"73816a9b-4677-4cb7-be74-67895d0dc2db","order_by":1,"name":"Uday P. Waghe","email":"","orcid":"","institution":"Yeshwantrao Chavan College of Engineering","correspondingAuthor":false,"prefix":"","firstName":"Uday","middleName":"P.","lastName":"Waghe","suffix":""},{"id":472222602,"identity":"4034c1e1-d45d-4732-8d43-e868781e8b7e","order_by":2,"name":"Prajakta Waghe","email":"","orcid":"","institution":"Yeshwantrao Chavan College of Engineering","correspondingAuthor":false,"prefix":"","firstName":"Prajakta","middleName":"","lastName":"Waghe","suffix":""}],"badges":[],"createdAt":"2025-06-10 12:08:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6862865/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6862865/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s40996-025-01979-z","type":"published","date":"2025-08-04T15:57:24+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":84912620,"identity":"87617760-174e-41e5-b138-0702513bc763","added_by":"auto","created_at":"2025-06-18 17:31:07","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":73156,"visible":true,"origin":"","legend":"\u003cp\u003eChemical composition of Fly ash \u0026amp; Glass Powder\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/2cbc36cd84234fb3a4cc0412.png"},{"id":84913671,"identity":"ffb031a6-e3c1-43d7-839e-54446fcd0bcc","added_by":"auto","created_at":"2025-06-18 17:39:07","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":75505,"visible":true,"origin":"","legend":"\u003cp\u003eChemical composition of Artificial Aggregate\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/5e64425e84dcd57ac92f0a90.png"},{"id":84912625,"identity":"cbc0795e-dd67-49c9-9b26-5d0e0578af32","added_by":"auto","created_at":"2025-06-18 17:31:07","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":386331,"visible":true,"origin":"","legend":"\u003cp\u003eMethodology Flow Chart for GPC ML Model\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/9309cfd0f7ad5e03d489cfc8.png"},{"id":84912658,"identity":"8cc616e9-453f-4465-b95b-a6a3d8c6254c","added_by":"auto","created_at":"2025-06-18 17:31:08","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":206690,"visible":true,"origin":"","legend":"\u003cp\u003eDurability metrics vs CDW Content\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/1b0fc34cd3630aa3dea15741.png"},{"id":84913672,"identity":"6b7c71ac-1af7-4226-86f5-2bcea580c0c6","added_by":"auto","created_at":"2025-06-18 17:39:07","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":286434,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance of GPC with varying CDW content\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/4e0fba92175d57a58506235c.png"},{"id":84913674,"identity":"9420c1e9-d670-4ebb-97fd-bb2b0069ba6a","added_by":"auto","created_at":"2025-06-18 17:39:07","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":138476,"visible":true,"origin":"","legend":"\u003cp\u003eML Model Performance\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/83495fc2ad1b612b86052df5.png"},{"id":84912630,"identity":"bd017da2-f6c7-4220-ab28-d9aa671b17fd","added_by":"auto","created_at":"2025-06-18 17:31:07","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":66209,"visible":true,"origin":"","legend":"\u003cp\u003eExperimental vs. Predicted Compressive Strength of Geopolymer Concrete\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/9e2f17d8b7e6471e921eea1f.png"},{"id":84912632,"identity":"bd1b68e6-1791-4b4c-b2cc-547beebd1ba2","added_by":"auto","created_at":"2025-06-18 17:31:07","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":59605,"visible":true,"origin":"","legend":"\u003cp\u003eExperimental vs. Predicted Residual Strength of Geopolymer Concrete\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/0353645e8915d14512713868.png"},{"id":84912639,"identity":"808c3952-777a-4911-8284-d3183f616fc4","added_by":"auto","created_at":"2025-06-18 17:31:07","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":66674,"visible":true,"origin":"","legend":"\u003cp\u003eExperimental vs. Predicted Mass Loss of Geopolymer Concrete\u003c/p\u003e","description":"","filename":"image9.png","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/6da3c8d5d78509c9f0038bcf.png"},{"id":84912634,"identity":"2b15e0ac-1ecd-475c-a9d9-ce18ad90880d","added_by":"auto","created_at":"2025-06-18 17:31:07","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":90984,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP-Based Feature Importance in XGBoost Model\u003c/p\u003e","description":"","filename":"image10.png","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/00c8eb71071065afa5fc435d.png"},{"id":84912650,"identity":"9d53dc46-d4d3-4c81-a567-3646378789dd","added_by":"auto","created_at":"2025-06-18 17:31:07","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":165743,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP dependence plot for Durability\u003c/p\u003e","description":"","filename":"image11.png","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/7034f3e42b189b81f85b9ef5.png"},{"id":88814262,"identity":"f787db39-b326-4d54-b224-fe08ca994189","added_by":"auto","created_at":"2025-08-11 16:09:25","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2912146,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6862865/v1/e824b3c1-43aa-4a3f-8eef-983cb6d0d799.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning Prediction Model for Durability of Geopolymer Concrete with CDW and Artificial Aggregates under Harsh Environmental Exposure","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe construction sector today faces growing pressure to reduce its environmental impact and manage natural resources more responsibly. Traditional concrete, which relies heavily on Portland cement and natural aggregates, contributes significantly to carbon emissions and the depletion of raw materials [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. As a response, geopolymer concrete (GPC) has been proposed as a more sustainable alternative [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. It offers lower carbon emissions and the ability to incorporate waste materials from other industries. Among these, various types of waste considered for use as a natural fine aggregate and used in concrete mixtures, CDW has received the most attention [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. CDW containing crushed bricks may help improve the internal structure of the concrete and provide some pozzolanic activity. Sulphate attack, chloride attack and freeze-thaw cycles are key conditions to consider when planning for long-term durability [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAt almost the same moment, machine-learning software began to pop up in laboratory notebooks, advertising a chance to snap quick forecasts out of wobbly test data. Open a dashboard, feed in a few dozen measurements, and the models sift through the hidden connections, slashing that brutal grind of trial-and-error pouring. In concrete circles, engineers now lean on those calculations to pin down durability and tinker with mixture recipes in a fraction of the usual fuss [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eConcrete researchers have lately turned to machine learning to forecast how mixes perform, with geopolymer concrete drawing particular interest. Early experiments leaned on familiar tools such as linear regression or support-vector regression, but the field has quickly pivoted toward ensemble giants like random forest and XGBoost [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Those algorithms tackle the tangled, non-linear relationships in the data with far less fuss. Artificial Neural Networks (ANNs) have also shown promise in capturing intricate mix-performance relationships, albeit with trade-offs in interpretability. However, most prior studies focus on strength prediction alone and seldom integrate explainable AI techniques such as SHAP to assess feature importance under varied durability exposures. To address this gap, the present study evaluates five diverse ML models\u0026mdash;LR, SVR, RF, ANN, and XGBoost\u0026mdash;to predict compressive strength retention in CDW-based GPC exposed to harsh environments. The selection of models reflects a deliberate mix of linear, kernel-based, tree-based, and neural network methods to assess predictive performance and model transparency under different algorithmic paradigms [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis research seeks to assess the durability performance of CDW and artificial aggregate-based geopolymer concrete with different contents under severe exposure conditions. It also assesses the predictive capacity of different ML algorithms to compressive strength retention, with feature importance analysis based on SHAP values [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. The experimental and computational strategy seeks to inform the design of sustainable, durable GPC products that can perform under actual environmental conditions.\u003c/p\u003e \u003cp\u003eThe novelty of this study is that it has an integrated approach to addressing a critical research issue in GPC durability. Some studies have investigated the mechanical properties of GPC or addressed some environmental conditions, but limited research has considered the integration of recycled CDW fine aggregate and artificial coarse aggregates and with a full durability test under severe conditions such as sulphate attack, freeze\u0026ndash;thaw, and chloride penetration. This work fills this gap by presenting comprehensive experimental data on how these parameters interact with each other and further employs machine learning models, namely XGBoost with SHAP analysis, to predict durability and emphasise how each mix parameter influences it. The combination of experimental and predictive modelling approaches offers a feasible and innovative solution to the creation of more durable and sustainable GPC mixes. This work presents a new framework that facilitates the use of recycled materials in real-world applications, to the benefit of the environment and performance under harsh service conditions.\u003c/p\u003e"},{"header":"2. Literature Review","content":"\u003cp\u003eGeopolymer concrete (GPC) is being studied as a more environmentally friendly option compared to ordinary Portland cement (OPC), mainly because it can make use of industrial waste materials and helps reduce carbon emissions [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Materials like fly ash and ground glass powder are commonly used as sources of aluminosilicate, which, when mixed with alkaline solutions, react to form a dense and durable binder. Research has shown that adding glass powder can improve the concrete\u0026rsquo;s structure and increase its resistance to chemical attacks such as sulphate and chloride exposure [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. At the same time, interest in recycling has grown, especially with the push toward circular economy practices. This has led to more studies exploring how construction and demolition waste (CDW) can be reused in concrete production [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. CDW, particularly when composed of crushed bricks and old mortar, exhibits pozzolanic activity and filler effects that contribute to improved performance in geopolymer systems. Research has shown that CDW can enhance freeze-thaw resistance, reduce chloride permeability, and limit mass loss in sulphate-rich environments. However, many of these studies focus only on single exposure conditions or limited mechanical tests, without examining combined or long-term environmental effects that better represent field conditions summarised in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eRecent developments in machine learning (ML) have opened new opportunities in concrete research, especially in modelling compressive strength and mix optimisation [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Yet, the application of ML to predict durability parameters, such as strength retention after aggressive exposures, remains limited. Only a few studies have used SHAP analysis to clearly explain how different mix ingredients, such as the amount of Construction and Demolition Waste (CDW) or the activator-to-binder ratio, affect the performance of geopolymer concrete. This points to a noticeable gap in current research [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Most existing studies focus either on experimental testing or on machine learning models separately, with few combining both approaches under different durability conditions. This study addresses that limitation by evaluating how geopolymer concrete containing CDW and artificial aggregates performs when exposed to aggressive environments. Machine learning is used not only to predict how the material behaves, but also to help understand which mix design factors have the greatest effect on strength retention [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparative Review of Studies on Geopolymer Concrete (GPC) Durability and Machine Learning Applications\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReferences\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMaterials Used\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDurability Parameters Studied\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMachine Learning Application\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNovelty\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eIdentified Research Gap\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGlass powder, OPC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSulphate and chloride resistance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eComprehensive review of glass powder as a pozzolan\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo GPC integration or recycled aggregate study\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFly ash-based GPC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCarbonation, sulphate, acid\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDurability under chemical exposure\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eExcludes recycled materials and ML integration\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCDW aggregates, OPC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFreeze\u0026ndash;thaw, permeability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDurability of CDW concrete\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNot based on a geopolymer matrix\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRecycled aggregates\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eShrinkage, strength, and workability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePartial replacement in OPC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo ML or durability under aggressive exposures\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSilica fume, fly ash\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eChloride ingress, compressive strength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDurability with SCMs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo use of CDW or predictive modelling\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFly ash GPC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSulphate and acid resistance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGPC under aggressive media\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eExcludes recycled aggregates\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFly ash, CDW\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCompressive strength, abrasion\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCDW-based GPC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo chemical/environmental durability study\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFly ash GPC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCompressive strength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eANN, SVR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eApplied ML for strength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo durability or explainable AI tools\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOPC concrete\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCompressive strength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF, ANN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eML prediction model\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo focus on GPC or recycled materials\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGlass powder, metakaolin\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSulphate and acid\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBlending recycled glass with binders\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo ML or CDW incorporation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCementitious composites\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eStrength and durability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSHAP, LIME\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eExplainable ML use\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNot focused on GPC or recycled aggregates\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAAS\u0026thinsp;+\u0026thinsp;CDW\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCarbonation, RCPT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAlkali-activated concrete with CDW\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo ML application\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFly ash GPC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLong-term sulphate resistance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLong-term study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo AI or CDW inclusion\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCDW, fly ash, GGBS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMechanical \u0026amp; durability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eComprehensive CDW\u0026ndash;GPC assessment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLacks AI-based prediction tools\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThis Study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFly ash, glass powder, CDW, artificial aggregates\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSulphate, chloride, freeze\u0026ndash;thaw, strength retention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGBoost, SHAP, ANN, SVR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eIntegrates recycled CDW and explainable ML for harsh exposure\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eBridges experimental data with interpretable AI for sustainable GPC design\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"3. Materials and Methodology","content":"\u003cp\u003eIn this study, both lab testing and machine learning techniques were used to examine how durable geopolymer concrete (GPC) is when made with recycled construction and demolition waste (CDW) and artificial coarse aggregates. The testing process was designed to understand the specific role of recycled materials and to reflect how the concrete might perform over time in harsh environmental conditions.\u003c/p\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Materials\u003c/h2\u003e \u003cp\u003eThe binder used in the geopolymer mix was made by combining low-calcium Class F fly ash with ground waste glass. Both materials contain aluminosilicates, which react well with alkaline solutions and help form the structure of the hardened concrete. They were chosen because of their ability to improve strength and contribute to the chemical stability of the mix.\u003c/p\u003e \u003cp\u003eCDW was collected from demolition sites and comprised primarily of crushed brick, mortar residues, and aged concrete. After screening and crushing, CDW was graded to fine aggregate size (passing 4.75 mm) and used as a partial or full replacement for natural sand. Artificial coarse aggregates, manufactured through sintering lightweight materials, were used across all mixes to ensure consistency in bulk density and enhance internal curing and thermal insulation properties [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. An alkaline activator solution, prepared with sodium hydroxide (10M concentration) and sodium silicate (approximately 2.0), served as the chemical trigger for the polymerisation process. The activator-to-binder (A/B) ratio was maintained at 0.45 throughout the study to ensure comparability across mixes. The oxide composition of both fly ash and glass powder, dominated by SiO₂, Al₂O₃, and minor CaO, was verified using X-ray fluorescence (XRF). Their reactivity contributes significantly to the formation of N-A-S-H gels, imparting mechanical and chemical stability under environmental stressors. The chemical composition of the materials used is as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e \u003cb\u003e\u0026amp;\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The research methodology is as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Mix Design\u003c/h2\u003e \u003cp\u003eThree distinct concrete mixtures were formulated to investigate the effect of CDW replacement levels, as shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The proportions of fly ash and glass powder were held constant in all mixes to ensure that observed differences in performance could be attributed solely to CDW content. The mixing procedure involved dry blending of binders and aggregates, followed by gradual addition of the activator solution and thorough mechanical mixing to achieve homogeneity.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDifferent Mix Samples\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMix ID\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCDW Content (% of Fine Aggregate)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFly Ash (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGlass Powder (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eActivator/Binder Ratio\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eArtificial Coarse Aggregate (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGPC-0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGPC-50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGPC-100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Sample Preparation and Curing\u003c/h2\u003e \u003cp\u003eCylindrical and prismatic specimens were cast using standard moulds (100 mm \u0026times; 200 mm for cylinders and 100 mm \u0026times; 100 mm \u0026times; 500 mm for beams). All samples were demoulded after 24 hours and cured under ambient laboratory conditions (23\u0026thinsp;\u0026plusmn;\u0026thinsp;2\u0026deg;C, 65% RH) for 28 days to simulate in-situ curing scenarios without artificial heat treatment.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Durability Testing Procedures\u003c/h2\u003e \u003cp\u003eTo evaluate long-term performance, specimens were subjected to the following durability assessments. Immersion in a 5% Na₂SO₄ solution was conducted for up to 90 days, with periodic mass loss measurements at 28, 56, and 90 days. Samples underwent 300 cycles of freezing (-18\u0026deg;C) and thawing (+\u0026thinsp;4\u0026deg;C) as per ASTM C666. Surface degradation was quantified by measuring surface scaling thickness. The Rapid Chloride Permeability Test (RCPT) was conducted under ASTM C1202, where the total charge passed (in Coulombs) was used to classify permeability levels. Post-exposure compressive strength tests were carried out to calculate strength retention ratios, offering a composite indicator of chemical and mechanical integrity [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Machine Learning Modelling\u003c/h2\u003e \u003cp\u003eA structured dataset incorporated key variables like CDW percentage, A/B ratio, exposure type, and measured durability outcomes. Five regression algorithms were deployed for strength retention prediction are Linear Regression (LR), Support Vector Regression (SVR), Random Forest (RF), XGBoost, and Artificial Neural Networks (ANN). To measure how well the models performed, their results were compared using standard error metrics: R\u0026sup2;, Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). Among the models tested, XGBoost gave the most accurate predictions. This is likely because of how it builds trees step by step to reduce errors and avoid overfitting. The machine learning models were used to predict how the concrete would perform under tough environmental conditions. The focus was on estimating compressive strength, weight loss, and how much strength remained after exposure, using information about the mix design and type of exposure.\u003c/p\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e3.5.1 Data Preprocessing\u003c/h2\u003e \u003cp\u003eBefore training the machine learning models, the experimental data were checked and organised. The inputs included details like how much fly ash and glass powder were used, the concentration of sodium hydroxide, the ratio of sodium silicate to sodium hydroxide, the amount of artificial and recycled aggregates, curing temperature, and how long the samples were exposed. The outputs being predicted were compressive strength (in MPa), percentage of weight loss, and strength remaining after exposure. Any errors or missing values in the data were removed to ensure consistency. Using a common scaling technique, all input values were made to fall within the same range, from 0 to 1, to facilitate comparison of the variables. To prevent bias during model training, this was required because variables with wider numeric ranges might\u003c/p\u003e \u003cp\u003eThe data used in this study came from a comprehensive experimental program that included 90 distinct samples produced by combining three mix designs (GPC-0, GPC-50, and GPC-100) and several exposure times under freeze-thaw, sulphate, and chloride conditions. Input features like CDW content, fly ash percentage, glass powder percentage, activator-to-binder (A/B) ratio, curing regime, and exposure type were included in every sample. Compressive strength, residual strength, and mass loss following environmental exposure were the output variables. Based on previous research and SHAP analysis, these inputs were chosen because of their known impact on durability.\u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section4\"\u003e \u003ch2\u003e3.5.1.1 Data Cleaning and Scaling\u003c/h2\u003e \u003cp\u003eAll input features were normalised to a 0\u0026ndash;1 range using Min-Max scaling before the model was trained. By taking this step, the learning process is prevented from being dominated by features with higher numerical values, like compressive strength or exposure duration. Because of the controlled experimental conditions, there were no missing or unusual data points. To ensure compatibility with all regression models, one-hot encoding was used to encode categorical variables, such as exposure type.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e \u003ch2\u003e3.5.2 Model Selection and Hyperparameter Tuning\u003c/h2\u003e \u003cp\u003eGradient Boosting Regressor (GBR), Random Forest Regressor (RFR), and Decision Tree Regressor (DTR) were the models utilised in this investigation. These models were chosen because of their resilience to outliers and capacity to manage non-linear relationships. Optimised hyperparameters were used during training for each model to increase accuracy and avoid overfitting. Grid search and five-fold cross-validation were used to adjust the hyperparameters. Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e displays the chosen hyperparameters.\u003c/p\u003e \u003cp\u003eUsing a grid search approach in conjunction with five-fold cross-validation, hyperparameter tuning was carried out to guarantee the best possible predictive accuracy across models. To determine which configuration produced the lowest error on validation sets, this method entailed methodically testing combinations of parameter values. Each model type had a different search range. No tuning was necessary for Linear Regression (LR). Support Vector Regression (SVR) was tuned for kernel type (linear, polynomial, RBF), regularisation parameter C (range: 0.1\u0026ndash;100), and epsilon margin (0.01\u0026ndash;1.0). Random Forest (RF) was tuned for the number of estimators (50\u0026ndash;200), maximum depth (4\u0026ndash;12), and minimum samples per split (2\u0026ndash;5). For XGBoost, the learning rate (0.01\u0026ndash;0.2), maximum depth (3\u0026ndash;10), and number of estimators (50\u0026ndash;200) were explored. Artificial Neural Network (ANN) tuning involved the number of hidden layers (1\u0026ndash;3), neurons per layer (16\u0026ndash;64), learning rate (0.001\u0026ndash;0.01), and batch size (8\u0026ndash;32). These parameters were selected based on prior studies in concrete prediction tasks and refined using experimental error analysis.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eOptimised Hyperparameters for Each ML Model\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eKey Hyperparameters\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDecision Tree Regressor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMax depth\u0026thinsp;=\u0026thinsp;6, Min samples split\u0026thinsp;=\u0026thinsp;4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest Regressor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEstimators\u0026thinsp;=\u0026thinsp;100, Max depth\u0026thinsp;=\u0026thinsp;8, Min samples split\u0026thinsp;=\u0026thinsp;3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGradient Boosting Regressor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLearning rate\u0026thinsp;=\u0026thinsp;0.1, Estimators\u0026thinsp;=\u0026thinsp;150, Max depth\u0026thinsp;=\u0026thinsp;5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eDurability prediction model selection was performed in two stages to determine which machine learning model would be the best fit. Initially, in the first stage, five most popular regression models, specifically, Linear Regression, Support Vector Regression (SVR), Random Forest, XGBoost, and Artificial Neural Networks (ANN) were fitted to the standard settings. This comparative study at the first stage helped to find out which methods were consistently accurate with the lowest error. Then, out of all the rest models, Random Forest and XGBoost exhibited the highest degrees of protecting data from complexities concerning the error rate and were thus the most acceptable.\u003c/p\u003e \u003cp\u003eAt this further stage, three of the tree-based models: Decision Tree Regressor, Random Forest Regressor, and Gradient Boosting Regressor were thoroughly tuned in their parameters to identify the best performing model. The models were selected for their ability to capture non-linear patterns and for their increased model training control. Their parameters were tuned via grid search and cross-validation, and the final configurations are provided in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Among those, XGBoost has the highest scores, so further investigation using SHAP will be conducted to make the input attribute effect clearer for the durability prediction model.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003ch2\u003e3.5.3 Cross-Validation Strategy\u003c/h2\u003e \u003cp\u003eTo check how well the models could make predictions on new data, a five-fold cross-validation method was used. The dataset was randomly split into five parts. The model was trained in four parts per round, and it was tested in the fifth part. In order to test every component of the data once, this procedure was carried out five times. The average of the five rounds was used to determine the final outcomes. This method reduces the likelihood that the model will simply memorise the training set and helps demonstrate how the model performs on data it hasn't seen before.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.6 SHAP-Based Feature Importance\u003c/h2\u003e \u003cp\u003eSHAP analysis was used to gain a better understanding of the factors that affected the machine learning predictions. This method demonstrated the relative contributions of each input variable to the model's output. It was discovered that the activator-to-binder ratio and the quantity of CDW had the biggest effects on the retention of compressive strength. The SHAP results helped to clarify how the XGBoost model arrived at its conclusions and were in good agreement with experimental trends. The biggest impact was from CDW content. As CDW replacement rose, strength retention generally improved, particularly up to roughly 70%. Following that, the benefit began to plateau or decline slightly, perhaps as a result of modifications to the internal structure or mix workability. The next most crucial element was the activator-to-binder ratio. A more compact and long-lasting concrete was produced when this ratio was balanced because it enhanced the chemical reaction between the binder and the activator. Another important factor was the kind of environmental exposure. Chloride was the exposure type that weakened the concrete the most, followed by sulphate and freeze-thaw. This is probably because chloride ions are highly reactive and can degrade the geopolymer matrix more forcefully than other agents.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Results and Discussion","content":"\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Experimental Durability Performance\u003c/h2\u003e \u003cp\u003eThe durability assessment of geopolymer concrete (GPC) incorporating varying proportions of construction and demolition waste (CDW) and artificial aggregates revealed clear performance trends under aggressive environmental exposures \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e\u003cb\u003e).\u003c/b\u003e As the CDW content increased, the mass loss after 90 days of sulphate exposure decreased notably, from 6.5% at 0% CDW to 3.7% at 100% CDW. This trend signifies enhanced sulphate resistance, likely due to the absence of Ca (OH)₂ in GPC and the pozzolanic contribution of crushed brick in CDW. Surface scaling under freeze-thaw cycles also diminished with higher CDW content, dropping from 2.7 mm to 1.4 mm. The enhanced freeze-thaw resistance can be attributed to the lower permeability and refined pore structure imparted by recycled fines and the lightweight artificial coarse aggregates.\u003c/p\u003e \u003cp\u003eStrength retention showed a positive correlation with CDW content. Samples with 70\u0026ndash;100% CDW retained up to 88% of their original compressive strength after sulphate attack, compared to just 72% in CDW-free mixes. Additionally, Rapid Chloride Penetration Test (RCPT) results indicated a reduction in chloride ion permeability, with charge passed decreasing from 1400 Coulombs (moderate permeability) at 0% CDW to 850 Coulombs (low permeability) at 100% CDW.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec16\" class=\"Section3\"\u003e \u003ch2\u003e4.1.1 Mass Loss under Sulphate Attack\u003c/h2\u003e \u003cp\u003eAfter 90 days of exposure to a 5% Na\u003csub\u003e₂\u003c/sub\u003eSO₄ solution, mass loss exhibited a clear declining trend with increasing CDW content. Specifically, the control mixture with 0% CDW lost approximately 6.5% of its mass, whereas the mixture composed entirely of CDW lost only about 3.7%. The intermediate mixture with 50% CDW lost about 5.0% over the same period. For example, at 28 days, the mass losses were 2.8% (0% CDW), 2.2% (50% CDW), and 1.7% (100% CDW). At 56 days, the corresponding losses were 5.2%, 4.1%, and 3.0%, respectively. These results are shown in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, that incorporating CDW substantially mitigates the destructive effects of sulphate attack on the material. This improvement may be attributed to the consumption of portlandite (Ca (OH)₂) and the resultant densification of the microstructure, thereby reducing the material\u0026rsquo;s susceptibility to sulphate-induced deterioration.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMass Loss of Geopolymer Concrete Under Sulphate Attack at Different CDW Replacement Levels Over Time\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExposure Time (Days)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMass Loss 0% CDW (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMass Loss 50% CDW (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMass Loss 100% CDW (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e6.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section3\"\u003e \u003ch2\u003e4.1.2 Surface Scaling under Freeze-Thaw Cycles\u003c/h2\u003e \u003cp\u003eFollowing 300 cycles of freezing and thawing, the extent of surface scaling was observed to decrease with increasing CDW content. For instance, the control specimens with no CDW exhibited severe scaling of approximately 2.7 mm, whereas those composed entirely of CDW showed only minor scaling of around 1.4 mm. This enhancement in freeze-thaw resistance can be attributed to reduced capillary porosity and the densification of the microstructure due to pozzolanic reactions contributed by the recycled brick fines. Although brick fines are hydrophilic and naturally absorb more water, their fine particle size and reactive silica-alumina content can help fill voids and enhance gel formation within the matrix. This leads to reduced interconnected capillary pores, which limit free water movement and mitigate internal pressure during freezing cycles introduced by the recycled brick fines. These factors inhibit water ingress and ice formation within the pore structure, thereby mitigating surface damage under cyclic freeze-thaw conditions.\u003c/p\u003e \u003cp\u003eAlthough both CDW and artificial lightweight aggregates are known to have higher intrinsic porosity and water absorption compared to natural aggregates, their influence on freeze\u0026ndash;thaw resistance is more nuanced. Several studies [Silva et al., 2021; Li et al., 2018] have shown that when used in geopolymer systems, the enhanced microstructural bonding between alkali-activated binders and recycled brick particles can reduce capillary pore continuity, thereby limiting water ingress and internal pressure build-up during freeze\u0026ndash;thaw cycles. Moreover, the internal curing effect of lightweight aggregates can help buffer moisture fluctuations by gradually releasing absorbed water, reducing pore saturation levels during freezing events [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. As a result, the apparent permeability may not fully reflect the effective transport pathways relevant to freeze\u0026ndash;thaw deterioration. Thus, despite higher absorption values, the observed surface scaling reduction can be attributed to improved pore structure refinement and internal moisture regulation mechanisms, consistent with findings from recent geopolymer durability research [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e \u003ch2\u003e4.1.3 Strength Retention\u003c/h2\u003e \u003cp\u003eThe retention of compressive strength after sulphate exposure was found to be highly dependent on the level of CDW replacement. Geopolymer concrete mixtures containing 70\u0026ndash;100% CDW retained approximately 85\u0026ndash;88% of their original compressive strength, whereas the CDW-free mixture retained only about 72%. These observations imply that higher CDW content leads to a geopolymer matrix that is chemically more stable and mechanically more resilient under aggressive sulphate attack shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. The findings indicate that the inclusion of recycled construction waste markedly bolsters the material\u0026rsquo;s resistance to strength loss in chemically aggressive environments.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section3\"\u003e \u003ch2\u003e4.1.4 Chloride Ion Penetrability (RCPT)\u003c/h2\u003e \u003cp\u003eThe rapid chloride permeability test (RCPT) results indicate that chloride ion permeability decreased markedly as CDW content increased. In practical terms, the 0% CDW mixture allowed a charge of about 1400 Coulombs to pass (classified as moderate permeability by ASTM C1202), whereas the 100% CDW mixture allowed only 850 Coulombs (classified as low permeability). This significant reduction in permeability is attributed to the formation of a denser geopolymer matrix, resulting from the pozzolanic reactivity and filler effects of the recycled brick materials. Incorporating CDW into the geopolymer matrix substantially hinders the penetration of chloride ions by creating a more compact pore structure.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Machine Learning Model Prediction\u003c/h2\u003e \u003cp\u003eTo assess the predictive reliability of different machine learning (ML) models for estimating compressive strength retention in geopolymer concrete, five algorithms were trained and validated using experimental data. The models included Linear Regression, Random Forest, Support Vector Regression (SVR), XGBoost, and Artificial Neural Networks (ANN). Their performances were compared using three widely recognised statistical metrics: The coefficient of determination (R\u0026sup2;), mean absolute error (MAE), and root mean square error (RMSE), as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e \u003cb\u003e\u0026amp;\u003c/b\u003e Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe tested models, XGBoost, showed the highest predictive accuracy, achieving an R\u0026sup2; value of 0.96, coupled with a low MAE of 1.1 and RMSE of 1.6. These results indicate its exceptional ability to learn and generalise from complex patterns present in the input variables, including CDW content, activator ratios, and environmental exposure types.\u003c/p\u003e \u003cp\u003eThe Random Forest and ANN models also showed good results, with R\u0026sup2; of 0.92 and 0.94. The models are also able to handle non-linear relationships. The Random Forest and ANN models are less accurate than the XGBoost model. The Random Forest model is also characterized by a high level of interpretability and robustness to overfitting. The ANN model is rather flexible and may suffer from variations in the training process. The Support Vector Regression model has R\u0026sup2; \u0026lt;0.88, and its performance is less effective compared to other models. In particular, it was difficult to set the hyperparameters in the case of this model. Thus, the model is sensitive to the features of the dataset. The Linear Regression model demonstrated the lowest level of accuracy. Although this model also has R\u0026sup2; of 0.78, it is not able to find dependencies in the case of non-linear data. This model cannot be used for this dataset. The MAE and RMSE values are 3.2 and 4.8, respectively.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo assess the predictive accuracy of the machine learning models, standard statistical metrics were employed. The coefficient of determination (R\u0026sup2;) measures how well the model explains variance in the observed data. MAE quantifies the average magnitude of errors, while RMSE penalizes larger deviations more heavily.\u003c/p\u003e \u003cp\u003eCoefficient of Determination (R\u0026sup2;)\u0026thinsp;=\u0026thinsp;1 - Σ(yi - ŷi)\u0026sup2; / Σ(yi - ȳ)\u0026sup2; Eq.\u0026nbsp;(1)\u003c/p\u003e \u003cp\u003eMean Absolute Error (MAE) = (1/n) Σ|yi - ŷi| Eq.\u0026nbsp;(2)\u003c/p\u003e \u003cp\u003eRoot Mean Square Error (RMSE)\u0026thinsp;=\u0026thinsp;sqrt((1/n) Σ(yi - ŷi)\u0026sup2;) Eq.\u0026nbsp;(3)\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eModel Performance Comparison\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eR\u0026sup2; (Strength)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMAE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRMSE\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLinear Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSupport Vector Regression (SVR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eArtificial Neural Network (ANN)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTo improve the predictive accuracy, each machine learning model's hyperparameters were adjusted. Because of their capacity to handle intricate non-linear patterns in data while simultaneously lowering overfitting rates, the Random Forest and XGBoost models distinguished themselves with their robust and consistent performance. To increase the accuracy of the XGBoost model, important parameters like learning rate, tree depth, and the number of estimators were changed. Even when applied to previously untested data, these modifications improved the model's ability to predict compressive strength. The model's stability was examined using cross-validation, which made sure it functioned well in various dataset sections.\u003c/p\u003e \u003cp\u003eEven though XGBoost produced the most accurate results, some people find it challenging to understand. To solve this, the model's prediction process was examined using SHAP analysis (described in Section \u003cspan refid=\"Sec22\" class=\"InternalRef\"\u003e4.3\u003c/span\u003e). The activator-to-binder ratio, the type of environmental exposure, and the percentage of construction and demolition waste (CDW) were among the most crucial input variables that were identified with the aid of this technique. These results improved mix design choices in geopolymer concrete and helped explain the model's decisions. Future Directions: In the future, it may be useful to explore hybrid models that combine different machine learning techniques to improve predictive accuracy. For instance, stacking or assembling XGBoost with Random Forest could provide a more robust prediction framework. Furthermore, integrating explainability techniques such as SHAP with machine learning models will continue to support more transparent and data-driven decisions in geopolymer concrete mix design.\u003c/p\u003e \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e \u003ch2\u003e4.2.1 Comparison Between Experimental Results and ML Predictions (GBR Model)\u003c/h2\u003e \u003cp\u003eTo further verify the performance of the Gradient Boosting model, predicted outputs were compared with experimental results for compressive strength (CS), mass loss (ML), and residual strength (RS) under harsh exposure conditions. The results are presented in Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, where mix proportions follow the same categorisation as described in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe machine learning predictions showed excellent agreement with the experimental results across all mix types (GPC-0, GPC-50, and GPC-100). The slight deviations were within acceptable margins, confirming the robustness of the model. Furthermore, the effectiveness of the model is visually reinforced in Figs.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, \u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e \u0026amp; Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e, where predictions closely follow the experimental trend line\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eExperimental vs ML Prediction Results for Different Mix Proportions\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"12\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c12\" colnum=\"12\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMix ID\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCDW (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFly Ash (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGlass Powder (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eActivator/Binder Ratio\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eArtificial Aggregate (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eCS Exp (MPa)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eCS Pred (MPa)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eML Exp (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eML Pred (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003eRS Exp (MPa)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c12\"\u003e \u003cp\u003eRS Pred (MPa)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGPC-0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e42.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e41.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e5.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e5.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e35.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e34.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGPC-50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e38.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e6.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e31.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGPC-100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e46.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e45.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e4.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e4.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e37.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e36.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e presents a comparative analysis of experimental and machine learning (ML)-predicted results for three different geopolymer concrete (GPC) mix designs incorporating varying percentages of construction and demolition waste (CDW). The mix IDs\u0026mdash;GPC-0, GPC-50, and GPC-100\u0026mdash;denote 0%, 50%, and 100% replacement of natural fine aggregates with CDW, respectively, while maintaining consistent proportions of fly ash (70%) and glass powder (30%) as binders, an activator-to-binder ratio of 0.45, and a fixed artificial aggregate content of 100%.\u003c/p\u003e \u003cp\u003eThe compressive strength (CS) values predicted by the machine learning model were very close to the results from the lab tests. For example, in the GPC-0 mix, the measured strength was 42.5 MPa, and the model gave a value of 41.9 MPa. Similar agreement was seen in GPC-50 and GPC-100, where the predicted strengths were 39.0 MPa and 45.7 MPa, compared to experimental values of 38.4 MPa and 46.2 MPa. This shows that the model worked well across different levels of CDW replacement and was able to make accurate predictions for various mix designs.\u003c/p\u003e \u003cp\u003eThe model demonstrated good performance for mass loss (ML) under aggressive conditions. The largest discrepancy between expected and actual values was 0.3%, indicating that the model did a good job of capturing the effects of environmental exposure. For example, the model predicted an ML of 4.6% in the GPC-100 mix, while the measured ML was 4.3%. Given that safety margins are frequently preferred in real-world scenarios, this slight overestimation might be helpful.\u003c/p\u003e \u003cp\u003eThe same pattern was also seen in the residual strength (RS) predictions following exposure. The experimental value of RS for GPC-0 was 35.6 MPa, a mere 2.3% difference from the predicted value of 34.8 MPa. For other mixes, similar outcomes were obtained, indicating that the model can accurately predict strength and durability performance following exposure.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAcross various geopolymer concrete mixes, the machine learning model performed consistently. The model can be used to estimate important properties of concretes made with recycled and artificial materials, according to the slight variations between the expected and actual results. When designing mixes that must satisfy strength and durability requirements in sustainable construction, this type of prediction is helpful.\u003c/p\u003e \u003cp\u003eWhen predicting the durability of geopolymer concrete containing CDW and artificial aggregates, XGBoost produced the most accurate results out of all the models that were tested. The operation of XGBoost is responsible for this high accuracy. It gradually improves predictions, learning from past errors to make better choices in subsequent iterations. Additionally, it has built-in controls to prevent overfitting by preventing the model from becoming overly complex. XGBoost can automatically determine which input features are most important, handle missing values, and handle both simple and complex input patterns. Because of these characteristics, it works particularly well when working with concrete data, where there are frequently intricate and non-linear relationships between ingredients and performance.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e4.3 SHAP-Based Feature Importance Analysis\u003c/h2\u003e \u003cp\u003eSHAP analysis was used to improve the XGBoost model's predictions' interpretability. This method revealed the underlying relationships between mix design parameters and durability outcomes by quantifying the individual contributions of each input variable to the prediction of compressive strength retention. The most important factor influencing the model's output among the input features was the CDW content. It exerted a strong positive influence on durability, particularly up to a replacement level of approximately 70%, beyond which the marginal gains diminished. This plateau effect suggests that while CDW can enhance the packing density and contribute to matrix densification at moderate levels, excessive substitution may introduce variability or adverse effects on matrix integrity.\u003c/p\u003e \u003cp\u003eThe activator-to-binder ratio was identified as the second most influential parameter. Its prominence underscores its role in governing the dissolution rate of aluminosilicates and subsequent geo-polymerisation. Optimal ratios were found to promote better gel formation and pore structure refinement, thereby enhancing resistance to environmental degradation. The type of exposure condition, whether chloride, sulphate, or freeze\u0026ndash;thaw, also significantly influenced strength retention. Among these, chloride attack was the most detrimental, followed by sulphate and freeze\u0026ndash;thaw exposure. The SHAP-based insights corroborate the experimental observations and reinforce the importance of balanced mix design. Specifically, they highlight the need to optimise CDW usage and alkali activator chemistry to develop durable, low-carbon concretes capable of withstanding harsh service environments. This type of feature-level explainability is critical for promoting the use of smart, data-driven mix design strategies in sustainable infrastructure development.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe bar chart titled \"SHAP-Based Feature Importance in XGBoost Model\" illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e, shows that the different input variables contribute to predicting the compressive strength retention of geopolymer concrete under harsh environmental conditions. Each bar shows the mean absolute SHAP value for a feature, indicating its average effect on the model\u0026rsquo;s output across the dataset. Among all the input features, the content of Construction and Demolition Waste (CDW) had the highest SHAP value (0.35), showing that it had the strongest effect on durability, particularly when used in suitable amounts. With a SHAP value of 0.25, the Activator-to-Binder Ratio followed, as it is crucial to the way the chemical reactions occur during the formation of geopolymers. Third place went to Exposure Type (0.20), which illustrates how various environmental factors, like sulphate, chloride, or freeze-thaw cycles, can impact the performance of concrete.\u003c/p\u003e \u003cp\u003eThe curing regime contributed to the early strength development and long-term stability of the geopolymer mix despite having a lower SHAP value (0.10). Fly ash (0.04) and glass powder (0.06) had less of an impact but were still involved in the concrete's overall behavior. This kind of analysis can be helpful in enhancing mix designs by highlighting the mix components that have the biggest effects on strength retention. These findings also lend credence to sensible choices that strike a balance between sustainability and durability in practical applications.\u003c/p\u003e \u003cp\u003eAccording to the SHAP analysis, one of the key variables influencing durability predictions was the sodium-to-aluminum (Na/Al) molar ratio. This outcome is in good agreement with current knowledge of geopolymer chemistry. To create a dense aluminosilicate gel that increases strength and resistance to chemical attack, a balanced Na/Al ratio is required. Excessive sodium levels can cause surface problems like efflorescence and leave behind unreacted alkalis. On the other hand, too little sodium may result in incomplete polymerization and a weaker internal structure. The type and amount of artificial aggregate also ranked highly, likely due to their effect on how pores form and how well the aggregate bonds with the binder.\u003c/p\u003e \u003cp\u003eA SHAP dependence plot for the Na/Al ratio showed that its influence was not linear. The plot revealed a peak effect on durability when the ratio was between about 0.9 and 1.2. Outside this range, the predicted benefit dropped off. This trend reflects how important it is to maintain the right balance of alkalis. If the sodium content is too high, the gel may become unstable, allowing ions to move more freely and possibly leading to cracks from repeated wetting and drying. If it\u0026rsquo;s too low, the geopolymerization reaction may not fully develop, resulting in a porous and weaker material. This type of model-based insight helps link prediction outcomes to actual material behavior and can support better decision-making in concrete design.\u003c/p\u003e \u003cp\u003eThe SHAP dependence plot for artificial fine aggregate replacement showed that durability improved as the replacement level increased, especially up to around 50%. After that point, the improvement slowed or slightly decreased. This pattern suggests that replacing natural sand with artificial fine aggregates from construction and demolition waste can help improve the internal structure of the concrete. A moderate amount of replacement appears to fill voids more effectively and reduce the number of tiny pores that allow water and chemicals to move through the material, making it more resistant to environmental damage. The presence of angular particles with rough surfaces likely improves the interfacial transition zone (ITZ) between the binder and aggregate phases, contributing to superior mechanical interlock and reduced permeability. However, excessive replacement may introduce inconsistencies in particle size distribution or residual contaminants, which could compromise homogeneity and lead to localized weakness. Thus, the SHAP-based interpretation highlights the importance of optimizing the aggregate replacement ratio to balance sustainability goals with long-term performance.\u003c/p\u003e \u003cp\u003e \u003cb\u003e4.3.1 Environmental Exposure Condition, Specifically the Number of Wet-Dry Cycles, As Interpreted from the SHAP Dependence Plot\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe SHAP dependence analysis for the number of wet-dry cycles revealed a consistently negative impact on the predicted durability index, with SHAP values decreasing steadily as the number of cycles increased. This outcome is consistent with known deterioration mechanisms in geopolymer concrete subjected to cyclic moisture fluctuations. Repeated wetting and drying accelerate the ingress of water and dissolved ions, which can cause internal microcracking, leaching of alkali ions, and progressive weakening of the aluminosilicate gel matrix. The SHAP dependence plot thus reflects the physical degradation pathways that manifest under prolonged environmental stress. Notably, the rate of SHAP value decline appeared to accelerate beyond 20 cycles, suggesting a threshold after which damage mechanisms become more aggressive or irreversible. This insight underscores the model\u0026rsquo;s ability to capture not just statistical correlations but meaningful durability trends under real-world exposure, thereby enhancing its practical utility in performance-based mix design.\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003e4.3.2 Dependence of Model Predictions on Durability\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e illustrates the SHAP dependence plot for Durability, which emerged as the most influential feature in the XGBoost model. The plot demonstrates a clear positive correlation between the standardized durability values and the corresponding SHAP values. This indicates that as durability increases, the model's prediction for fatigue life also increases, signifying a strong and consistent contribution of durability to the output variable. The scatter pattern reveals a relatively linear relationship, suggesting minimal interaction effects within the observed range, and affirming that durability serves as a stable and independent predictor in the trained model.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn line with previous test results and engineering practice, the model indicates that longer service life is associated with higher durability values when the material is subjected to repeated loading. The distribution of the data points further demonstrates the model's adaptability to various scenarios. This demonstrates that the model is capturing relationships that make sense from a physical and engineering point of view and supports the notion that durability has a significant impact on how materials behave under fatigue. The comparison of different models is shown in Table\u0026nbsp;\u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of Machine Learning Models for Predicting Concrete Durability\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy / Reference\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eConcrete Type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eInput Features\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eML Model Used\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eModel Interpretability\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eReported Prediction Accuracy\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFly ash-based geopolymer concrete\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMix proportions, curing conditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLinear Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eModerate\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOPC concrete under sulphate exposure\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWater\u0026ndash;cement ratio, aggregate properties\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eModerate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRecycled aggregate concrete\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePorosity, slump, compressive strength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSupport Vector Machine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eModerate\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThis Study (2025)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGeopolymer concrete with CDW \u0026amp; AA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWater absorption, capillary uptake, acid resistance, mix ratios\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGBoost\u0026thinsp;+\u0026thinsp;SHAP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis study confirms that machine learning models can reliably predict the durability of geopolymer concrete made with CDW and artificial aggregates. Among the five models tested, XGBoost delivered the most accurate results, with an R\u0026sup2; value of 0.96, a mean absolute error (MAE) of 1.1, and a root mean square error (RMSE) of 1.6. These results closely matched the experimental data, such as compressive strength values of 42.5 MPa (experiment) vs. 41.9 MPa (prediction) for GPC-0, and 46.2 MPa vs. 45.7 MPa for GPC-100.\u003c/p\u003e \u003cp\u003eThe durability results also showed a clear benefit from using CDW. Strength retention after sulphate exposure reached up to 88% for high-CDW mixes, while chloride ion permeability dropped from 1400 to 850 Coulombs as CDW content increased. SHAP analysis helped explain the role of different variables, with CDW content and the activator-to-binder ratio having the greatest influence on durability.\u003c/p\u003e \u003cp\u003eOverall, combining experimental testing with interpretable ML tools offers a practical way to design more sustainable and long-lasting concrete mixes, particularly for structures exposed to aggressive conditions like sulphate, chloride, or freeze\u0026ndash;thaw environments.\u003c/p\u003e"},{"header":"6. Future Scope","content":"\u003cp\u003eBuilding upon the outcomes of this study, several directions are recommended for future research. First, while the current work focused on the durability of geopolymer concrete under sulphate, freeze\u0026ndash;thaw, and chloride exposures, future studies may consider additional degradation mechanisms such as carbonation, alkali\u0026ndash;silica reaction (ASR), and combined environmental attacks to assess long-term performance comprehensively. Furthermore, life cycle assessment (LCA) and cost-benefit analysis of CDW-based GPC mixtures would offer critical insights into their environmental and economic viability on a broader scale.\u003c/p\u003e \u003cp\u003eFrom a modelling perspective, integrating ensemble or hybrid machine learning approaches, such as model stacking or deep learning frameworks, may enhance predictive accuracy and generalisation across varied datasets. Additionally, incorporating real-time monitoring data from field applications could bridge the gap between laboratory-scale experimentation and practical implementation. Exploring explainable AI (XAI) tools beyond SHAP, such as LIME or counterfactual analysis, may also provide deeper interpretability of model predictions for decision-making in mix design.\u003c/p\u003e \u003cp\u003eLastly, scaling up the application of CDW-incorporated GPC in structural elements, pavements, or marine infrastructure, coupled with rigorous structural performance testing, will be critical for validating the suitability of this sustainable material in diverse civil engineering contexts. Such efforts will support the transition toward resilient, low-carbon construction practices aligned with global sustainability goals. Although the model achieved high accuracy, the dataset was limited to a specific set of materials and conditions. Future research could explore larger, multi-source datasets and additional environmental exposure types to enhance generalizability\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eGPC: Geopolymer Concrete, CDW: Construction and Demolition Waste, ML: Machine Learning, XGB / XGBoost: eXtreme Gradient Boosting, RF: Random Forest, ANN: Artificial Neural Network, SVR: Support Vector Regression, LR: Linear Regression, R\u0026sup2;: Coefficient of Determination, RMSE \u0026ndash; Root Mean Square Error, SHAP \u0026ndash; SHapley Additive exPlanations, NaOH: Sodium Hydroxide, Na\u003csub\u003e2\u003c/sub\u003eSiO\u003csub\u003e3\u003c/sub\u003e : Sodium Silicate, \u003cb\u003eA/B\u003c/b\u003e: Activator-to-Binder ratio,\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eConflicts of Interest:\u003c/h2\u003e \u003cp\u003eNo Conflicts of Interest.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAtul S. Kurzekar: Conceptualization, Methodology, Writing Original Draft, Software. Uday P. Waghe: Methodology, Investigation, Writing, Review \u0026amp; Editing. Prajakta Waghe: Formal Analysis, Validation, Resources.\u003c/p\u003e\u003ch2\u003eData Availability:\u003c/h2\u003e \u003cp\u003eThe data sets used and/or analysed during the current study are available from the Corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eMohajerani A et al (2019) Recycling waste materials in geopolymer concrete. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10098-018-01660-2\u003c/span\u003e\u003cspan address=\"10.1007/s10098-018-01660-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGunasekara C, Setunge S, Law DW, Willis N, Burt T (2018) Engineering Properties of Geopolymer Aggregate Concrete. J Mater Civ Eng 30(11). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1061/(asce)mt.1943-5533.0002501\u003c/span\u003e\u003cspan address=\"10.1061/(asce)mt.1943-5533.0002501\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChristensen TH, Birgisdottir H (2010) Recycling of Construction and Demolition Waste, in \u003cem\u003eSolid Waste Technology \u0026amp; Management\u003c/em\u003e, vol. 1. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/9780470666883.ch19\u003c/span\u003e\u003cspan address=\"10.1002/9780470666883.ch19\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo H et al (2018) Durability of recycled aggregate concrete \u0026ndash; A review. Cem Concr Compos 89. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cemconcomp.2018.03.008\u003c/span\u003e\u003cspan address=\"10.1016/j.cemconcomp.2018.03.008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJaniesch C, Zschech P, Heinrich K (2021) Machine learning and deep learning. Electron Markets 31(3). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s12525-021-00475-2\u003c/span\u003e\u003cspan address=\"10.1007/s12525-021-00475-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSarker IH (2021) Machine Learning: Algorithms, Real-World Applications and Research Directions. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s42979-021-00592-x\u003c/span\u003e\u003cspan address=\"10.1007/s42979-021-00592-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlomari Y, And\u0026oacute; M (2024) SHAP-based insights for aerospace PHM: Temporal feature importance, dependencies, robustness, and interaction analysis. Results Eng 21. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.rineng.2024.101834\u003c/span\u003e\u003cspan address=\"10.1016/j.rineng.2024.101834\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e8, Atul S, Kurzekar S, Kurzekar et al (2020) An Investigation on Structural Performance of Tall Building Embedded with Core, Outrigger System. Int J Mech Prod Eng Res Dev 10(3):2043\u0026ndash;2048. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.24247/ijmperdjun2020191\u003c/span\u003e\u003cspan address=\"10.24247/ijmperdjun2020191\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShivaprasad KN, Das BB (2018) Determination of optimized geopolymerization factors on the properties of pelletized fly ash aggregates. Constr Build Mater 163. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.conbuildmat.2017.12.038\u003c/span\u003e\u003cspan address=\"10.1016/j.conbuildmat.2017.12.038\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGerges NN, Issa CA, Fawaz SA, Jabbour J, Jreige J, Yacoub A (2018) Recycled Glass Concrete: Coarse and Fine Aggregates. Eur J Eng Technol Res 3(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.24018/ejeng.2018.3.1.533\u003c/span\u003e\u003cspan address=\"10.24018/ejeng.2018.3.1.533\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePatel S, Pansuria A, Shah V, Patel S (2014) Construction and Demolition Waste Recycling. IJIRST \u0026ndash;International J Innovative Res Sci Technology|, 1, 7\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMahesh B (2020) Machine Learning Algorithms - A Review. Int J Sci Res (IJSR) 9(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.21275/art20203995\u003c/span\u003e\u003cspan address=\"10.21275/art20203995\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePaleyes A, Urma RG, Lawrence ND (2022) Challenges in Deploying Machine Learning: A Survey of Case Studies. ACM Comput Surv 55(6). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1145/3533378\u003c/span\u003e\u003cspan address=\"10.1145/3533378\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRashidian-Dezfouli H, Rao Rangaraju P (2018) Evaluation of Selected Durability Properties of Portland Cement Concretes Containing Ground Glass Fiber as a Pozzolan, \u003cem\u003eTransportation Research Record: Journal of the Transportation Research Board\u003c/em\u003e, vol. 2672, no. 27, pp. 88\u0026ndash;98, Dec. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/0361198118773198\u003c/span\u003e\u003cspan address=\"10.1177/0361198118773198\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBernal SA, Rodr\u0026iacute;guez ED, Mej\u0026iacute;a R, de Guti\u0026eacute;rrez, Provis JL (2012) Performance of alkali-activated slag mortars exposed to acids, \u003cem\u003eJ Sustain Cem Based Mater\u003c/em\u003e, vol. 1, no. 3, pp. 138\u0026ndash;151, Sep. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/21650373.2012.747235\u003c/span\u003e\u003cspan address=\"10.1080/21650373.2012.747235\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePatil AS, Kurzekar AS, Fulzele P (2019) Behaviour of H sections with slender flange built upsection in pre engineering building. J Adv Res Dyn Control Syst, 11, no. 8 Special Issue\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePoon C-S, Chan D (May 2007) The use of recycled aggregate in concrete in Hong Kong. Resour Conserv Recycl 50(3):293\u0026ndash;305. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.resconrec.2006.06.005\u003c/span\u003e\u003cspan address=\"10.1016/j.resconrec.2006.06.005\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSiddique R, Khatib J, Kaur I (2008) Use of recycled plastic in concrete: A review. Waste Manag 28(10):1835\u0026ndash;1852. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.wasman.2007.09.011\u003c/span\u003e\u003cspan address=\"10.1016/j.wasman.2007.09.011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChokkalingam RB, Ganesan N (2017) A study on the strength development of geopolymer concrete using fly ashing fly ash. Int J Eng Technology(UAE) 6(4). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.14419/ijet.v6i4.7761\u003c/span\u003e\u003cspan address=\"10.14419/ijet.v6i4.7761\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSherwani AFH, Faraj R, Younis KH, Daraei A (2021) Strength, abrasion resistance and permeability of artificial fly-ash aggregate pervious concrete. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cscm.2021.e00502\u003c/span\u003e\u003cspan address=\"10.1016/j.cscm.2021.e00502\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNguyen-Sy T, Wakim J, To Q-D, Vu M-N, Nguyen T-D, Nguyen T-T (Nov. 2020) Predicting the compressive strength of concrete from its compositions and age using the extreme gradient boosting method. Constr Build Mater 260:119757. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.conbuildmat.2020.119757\u003c/span\u003e\u003cspan address=\"10.1016/j.conbuildmat.2020.119757\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWaghmare C et al (2025) Machine learning based prediction of compressive strength in roller compacted concrete: a comparative study with PDP analysis. Asian J Civil Eng. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s42107-025-01310-1\u003c/span\u003e\u003cspan address=\"10.1007/s42107-025-01310-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSingh H, Siddique R (Oct. 2023) Long term durability assessment of self-compacting concrete made with crushed recycled glass and metakaolin. Constr Build Mater 400:132656. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.conbuildmat.2023.132656\u003c/span\u003e\u003cspan address=\"10.1016/j.conbuildmat.2023.132656\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOkasha NM, Mirrashid M, Naderpour H, Ciftcioglu AO, Meddage DPP, Ezami N (2024) Machine learning approach to predict the mechanical properties of cementitious materials containing carbon nanotubes, \u003cem\u003eDevelopments in the Built Environment\u003c/em\u003e, vol. 19, p. 100494, Oct. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.dibe.2024.100494\u003c/span\u003e\u003cspan address=\"10.1016/j.dibe.2024.100494\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYaphary YL, Lau D, Sanchez F, Poon CS (May 2020) Effects of sodium/calcium cation exchange on the mechanical properties of calcium silicate hydrate (C-S-H). Constr Build Mater 243:118283. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.conbuildmat.2020.118283\u003c/span\u003e\u003cspan address=\"10.1016/j.conbuildmat.2020.118283\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBilir T, Gencel O, Topcu IB (2015) Properties of mortars with fly ash as fine aggregate. Constr Build Mater 93. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.conbuildmat.2015.05.095\u003c/span\u003e\u003cspan address=\"10.1016/j.conbuildmat.2015.05.095\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNaveena MP, Narayana G, Ravichandra R (2017) Study on strength and durability properties of fly ash aggregates concrete. Indian Concrete J, 91, 7\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKurzekar AS, Waghe U, Gawande D, Jaiswal A, Narkhede P (2025) Examining Fine Aggregate Properties from Dismantled Building Materials Waste: Towards Sustainable Construction Solutions in Environmental Sensing. 1\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-3-031-73816-6_1\u003c/span\u003e\u003cspan address=\"10.1007/978-3-031-73816-6_1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKurzekar AS, Waghe UP, Bhagat RM, Kherde YP (2024) Testing of mortar properties using dismantled building material and deteriorated bricks as substitution of natural fine aggregate for the reduction of cement. 070005. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1063/5.0240286\u003c/span\u003e\u003cspan address=\"10.1063/5.0240286\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJain KL, Sancheti G, Gupta LK (2020) Durability performance of waste granite and glass powder added concrete. Constr Build Mater 252. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.conbuildmat.2020.119075\u003c/span\u003e\u003cspan address=\"10.1016/j.conbuildmat.2020.119075\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaravanakumar P (2018) Strength and durability studies on geopolymer recycled aggregate concrete. Int J Eng Technology(UAE) 7(2). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.14419/ijet.v7i2.24.12087\u003c/span\u003e\u003cspan address=\"10.14419/ijet.v7i2.24.12087\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMasood B, Elahi A, Barbhuiya S, Ali B (2020) Mechanical and durability performance of recycled aggregate concrete incorporating low calcium bentonite. Constr Build Mater 237. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.conbuildmat.2019.117760\u003c/span\u003e\u003cspan address=\"10.1016/j.conbuildmat.2019.117760\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShaikh FUA (2016) Mechanical and durability properties of fly ash geopolymer concrete containing recycled coarse aggregates. Int J Sustainable Built Environ 5(2). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijsbe.2016.05.009\u003c/span\u003e\u003cspan address=\"10.1016/j.ijsbe.2016.05.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhmad A, Ahmad W, Aslam F, Joyklad P (Jun. 2022) Compressive strength prediction of fly ash-based geopolymer concrete via advanced machine learning techniques. Case Stud Constr Mater 16:e00840. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cscm.2021.e00840\u003c/span\u003e\u003cspan address=\"10.1016/j.cscm.2021.e00840\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eElahi MMA et al (Apr. 2021) Improving the sulfate attack resistance of concrete by using supplementary cementitious materials (SCMs): A review. Constr Build Mater 281:122628. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.conbuildmat.2021.122628\u003c/span\u003e\u003cspan address=\"10.1016/j.conbuildmat.2021.122628\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeng Y, Unluer C (Mar. 2023) Modeling the mechanical properties of recycled aggregate concrete using hybrid machine learning algorithms. Resour Conserv Recycl 190:106812. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.resconrec.2022.106812\u003c/span\u003e\u003cspan address=\"10.1016/j.resconrec.2022.106812\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"iranian-journal-of-science-and-technology-transactions-of-civil-engineering","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"istc","sideBox":"Learn more about [Iranian Journal of Science and Technology, Transactions of Civil Engineering](http://link.springer.com/journal/40996)","snPcode":"40996","submissionUrl":"https://submission.nature.com/new-submission/40996/3","title":"Iranian Journal of Science and Technology, Transactions of Civil Engineering","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Geopolymer concrete, Artificial aggregate, Machine Learning","lastPublishedDoi":"10.21203/rs.3.rs-6862865/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6862865/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis research explores how machine learning (ML) models can predict the durability of geopolymer concrete (GPC) produced using construction and demolition waste (CDW) as fine aggregate and artificial lightweight coarse aggregates. Durability was assessed under harsh exposure conditions, including sulphate attack, chloride penetration, and freeze\u0026ndash;thaw cycles. The study involved 90 experimental samples across varying CDW replacement levels. Five regression models, such as Linear Regression, Support Vector Regression, Random Forest, XGBoost, and Artificial Neural Networks, were trained to forecast strength retention based on mix composition and environmental exposure. Among them, XGBoost demonstrated the highest predictive accuracy (R\u0026sup2; = 0.96). SHAP analysis was used to explain model predictions and identify key influencing parameters, with CDW content and the activator-to-binder ratio emerging as critical factors. The findings show that moderate CDW incorporation enhances durability while reducing environmental impact, and that interpretable ML tools can assist in optimising mix designs for long-term performance in aggressive environments.\u003c/p\u003e","manuscriptTitle":"Machine Learning Prediction Model for Durability of Geopolymer Concrete with CDW and Artificial Aggregates under Harsh Environmental Exposure","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-18 17:31:02","doi":"10.21203/rs.3.rs-6862865/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-07-11T05:55:14+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-03T08:20:14+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"214846700463777440537154723853277103459","date":"2025-06-18T08:24:48+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-18T00:13:13+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-16T23:32:37+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"308309828923819014575277781300021933358","date":"2025-06-16T23:24:30+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"265734534443068732104516937659684540550","date":"2025-06-16T14:56:57+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"23384723368896350079357448637357970961","date":"2025-06-16T14:52:22+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-06-16T14:43:05+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-06-10T14:31:05+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-06-10T14:30:44+00:00","index":"","fulltext":""},{"type":"submitted","content":"Iranian Journal of Science and Technology, Transactions of Civil Engineering","date":"2025-06-10T11:55:16+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"iranian-journal-of-science-and-technology-transactions-of-civil-engineering","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"istc","sideBox":"Learn more about [Iranian Journal of Science and Technology, Transactions of Civil Engineering](http://link.springer.com/journal/40996)","snPcode":"40996","submissionUrl":"https://submission.nature.com/new-submission/40996/3","title":"Iranian Journal of Science and Technology, Transactions of Civil Engineering","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"fa489d42-fa62-417d-b843-2145adf16781","owner":[],"postedDate":"June 18th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-08-11T16:07:26+00:00","versionOfRecord":{"articleIdentity":"rs-6862865","link":"https://doi.org/10.1007/s40996-025-01979-z","journal":{"identity":"iranian-journal-of-science-and-technology-transactions-of-civil-engineering","isVorOnly":false,"title":"Iranian Journal of Science and Technology, Transactions of Civil Engineering"},"publishedOn":"2025-08-04 15:57:24","publishedOnDateReadable":"August 4th, 2025"},"versionCreatedAt":"2025-06-18 17:31:02","video":"","vorDoi":"10.1007/s40996-025-01979-z","vorDoiUrl":"https://doi.org/10.1007/s40996-025-01979-z","workflowStages":[]},"version":"v1","identity":"rs-6862865","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6862865","identity":"rs-6862865","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}