Machine Learning Approach for Mechanical Property Assessment of Industrial Waste-Filled Epoxy-Jute Composites

Machine Learning Approach for Mechanical Property Assessment of Industrial Waste-Filled Epoxy-Jute Composites | 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 Article Machine Learning Approach for Mechanical Property Assessment of Industrial Waste-Filled Epoxy-Jute Composites Abhilash Purohit, S. Sathees Kumar, Pravat Ranjan Pati, Arvind Kumar, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8903744/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 17 You are reading this latest preprint version Abstract This study investigates the development, characterization, and predictive modelling of sustainable epoxy composites reinforced with jute fiber and Linz-Donawitz (LD) sludge of industrial waste. Specimens were made using different LD sludge content (0–25 wt.%) and the constant jute fiber loading of 20 wt.%. The mechanical testing showed that the best composition was 60 wt.% epoxy, 20 wt.% jute, 20 wt.% LD sludge that demonstrated considerable gains in tensile strength (up to 61.84 MPa, an improvement of 28.8%), flexural strength (up to 31.81 MPa, an improvement of 41.8%), and impact strength (up to 18.026 kJ/m 2 ). After 20 wt.% sludge, the mechanical performance decreased because of interfacial defects. Machine learning models, namely Decision Tree, Random Forest, Gradient Boosting, and XGBoost were implemented in Google Colab to predict mechanical properties from compositional inputs. XGBoost showed better predictive performance as its error measures were close to zero (MAE = 0.0005 MPa tensile strength), validating its utility for inverse material design and the optimization of sustainable hybrid composites. The findings highlight a combined waste-valorisation and data-driven design approach that supports the development of sustainable composites for structural and industrial use. Physical sciences/Engineering Physical sciences/Materials science Composites Industrial Waste Jute Fiber Machine Learning Mechanical Characterization Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction The increased interest in sustainability and environmental responsibility in materials science has enhanced studies on eco-friendly composites that are reinforced with both natural fibers and industrial waste fillers [ 1 – 3 ]. Jute fiber (JF) is one natural fiber that has been revealed to be a promising reinforcement material because it is biodegradable, has low density, high specific strength, and is also cost-effective and thus suitable as reinforcement in composites that use polymer as matrix [ 4 – 6 ]. Simultaneously, the valorisation of industrial by-products like Linz-Donawitz (LD) sludge, which is an example of metallurgical waste product during the steel production, is a two-fold solution of disposing of waste and improving the composite quality [ 7 – 9 ]. Epoxy resins are widely used as a matrix material in composites due to their high mechanical strength, chemical resistance, and adhesive ability [ 10 – 12 ]. However, epoxy is a brittle material, has low impact strength and medium stiffness, which limit its application in the construction sector [ 13 , 14 ]. In order to overcome these shortcomings, fillers as well as the hybrid reinforcements have been extensively explored. Hybrid composites are those that have two or more different reinforcements, and they may have synergistic effects that enhance mechanical, thermal and physical performance [ 15 – 18 ]. The inclusion of LD sludge in the epoxy-jute composites is a novel method for developing sustainable hybrid materials. LD sludge may serve as a low-cost filler with improvements in hardness, density and mechanical strength due to dispersion strengthening and microstructural modification [ 19 ]. However, the mechanical performance of such composites is highly dependent on factors such as filler loading, interfacial adhesion, particle dispersion, and matrix-filler compatibility [ 20 ]. Purohit and Satapathy have investigated the mechanical, sliding wear [ 21 ] and erosion wear [ 22 ] behavior of LD sludge filled epoxy composites [ 23 ], and found that the addition of only particulate filler reduces the mechanical properties as a result of stress concentration. Abhilash Purohit et al. [ 24 ] examined epoxy composites reinforced with wood apple dust (WAD) and LinzDonawitz (LD) sludge and studied their mechanical and wear properties. The composite containing 5 wt.% WAD and 12 wt.% LD sludge demonstrated higher flexural (24%), compressive (9.88%), and tensile (30.2) strengths, and RSM analysis proved the high dependence of wear strength on filler additio. Purohit et al. [ 25 ] examined the mechanical and dry sliding wear behavior of epoxy composites reinforced with LD sludge, BF slag and LD slag. A Taguchi design was used to measure mechanical properties and wear performance, where LD sludge-filled epoxy composites were known to have better mechanical and wear properties, and their strength increased up to 15% in the case of epoxy composites with 16 wt.% of sludge fillers. The application of machine learning (ML) in materials science has revolutionized the design and optimization of novel composites because it has the capability to predict, with high precision and based on data, the mechanical characteristics of a given material by using compositional variables. Recent research shows that ensemble tree-based algorithms like Random Forest, Gradient Boosting, and XGBoost are better than traditional regression algorithms. Based on the study of Kumar et al. [ 26 ], the proposed study used machine learning to predict the publicly available natural fiber reinforced polymer (NFRP) datasets with the help of the following input parameters: epoxy content, density, and filler ratio. Random Forest was found to have the highest predictive power (R 2 = 0.92, MAE = 1.64 MPa) compared to XGBoost, Gradient Boosting and Polynomial Regression and was tested using five-fold cross-validation. Beldar and Kadbhane [ 27 ] developed predictive models of NaOH-treated jute fiber-reinforced glass epoxy composites. XGBoost demonstrated the highest performance with an R 2 of 0.9714 (tensile), 0.9803 (hardness) and 0.9718 (flexural) with the MSE values of 209.84, 52.02 and 127.71, respectively, due to regularization and tree pruning using five-fold cross-validation. The study by Milad et al. [ 28 ] involved XGBoost, multivariate adaptive regression spline (MARS), and random forest (RF) models in predicting the strain enhancement ratio of FRP composites based on experimental data from the literature. This paper investigated and compared the XGBoost, MARS and FRP models to predict the strain enhancement ratio in FRP composites using 729 experimental datasets provided by the literature. There were five input combinations formed due to the main material, strain, strength, and confinement parameters. All the models performed well, with MARS achieving the best accuracy, whereas RF also performed well with only strain properties. Based on these previous studies, it is noted that the wear resistance of the composites is enhanced when Linz Donawitz (LD) sludge because it is predominantly composed of iron oxides and silicon dioxide. The mechanical properties however, decrease with an increase in LD sludge percentage in the composite because of its irregular shape. This reducion in mechanical strength when a particulate filler is added can be avoided by adding another reinforcing phase in the form of a fiber [ 29 – 31 ]. According to these observations, to eliminate the decline in mechanical properties, the LD sludge was mixed in different proportions and strengthened using jute/ epoxy composites. The tensile, flexural and impact strengths were used to examine the mechanical properties. Nevertheless, machine learning models were used to confirm the findings of the experimental results. Materials and Methods Materials selection and fabrication method In this study, LY-556 epoxy (EP) was used as the matrix material along with the corresponding hardener. The epoxy and hardener were mixed in a 10:1 ratio as recommended by the supplier (Scientific supplier, Bhubaneswar, India). A natural fiber namely jute fiber (JF) was chosen to improve the mechanical properties of the composites. However, an industrial waste called LD sludge (LDS) was used to improve the hardness of the composites. LDS particles were collected from the Rourkela steel plant, Odisha, India, and sieved using a standard mesh to obtain a uniform particle size around 100 µm. The LDS particles were kept in an oven at 103°C for four hours to remove moisture. Hybrid EP-JF-LDS composites were fabricated by the hand-lay-up technique. Six specimens were prepared with different percentages of matrix and reinforcement phases, and the details are given in Table 1. The neat epoxy specimens (S0) were prepared to compare the properties of the other composites (S1, S2, S3, S4, and S5) with the neat epoxy. The designation of the specimens, along with the percentages of LDS and JF, is shown in Table 1. Table 1 Nomenclature and classification of composite specimens Specimen Designation EP-JF-LDS Composition S0 100 wt. % EP + 0 wt. % JF + 0 wt. % LDS S1 75 wt. % EP + 20 wt. % JF + 5 wt. % LDS S2 70 wt. % EP + 20 wt. % JF + 10 wt. % LDS S3 65 wt. % EP + 20 wt. % JF + 15 wt. % LDS S4 60 wt. % EP + 20 wt. % JF + 20 wt. % LDS S5 55 wt. % EP + 20 wt. % JF + 25 wt. % LDS Testing methodology and characterization of composite samples The tensile properties of the EP-JF-LDS hybrid composites were determined using the ASTM D3039 standard. Samples with 165 mm × 19 mm × 3.2 mm dimensions were prepared for the tensile tests. The accuracy of the test was within the range of ± 2%. Flexural strength of the EP-JF-LDS composites was determined as per the ASTM D790 [32] standard. Specimens with a gauge length of 51 mm and dimensions of 127 mm × 12.7 mm × 3.2 mm were prepared to carry out the flexural tests. The impact strength was measured in accordance with ASTM D256 using a pendulum impact tester. Samples with a 45° notch and 64 mm × 12.7 mm × 3.2 mm dimensions were prepared for the test [33, 34]. Three samples of the same composition were used for tensile, impact and flexural tests, and the average of the three readings was used to determine the final tensile strength value. ML models for mechanical properties estimation This study used four different machine learning regression models to estimate the mechanical properties, namely tensile, flexural and impact strengths, of epoxy-jute-LD sludge composites. The models chosen are Decision Tree Regression, Random Forest Regression, Gradient Boosting Regression and XGBoost (Extreme Gradient Boosting) Regression. These models have been selected to compare and evaluate their predictive power and generalization (A simple interpretable baseline (Decision Tree) to more complex ensemble approaches: Random Forest, Gradient Boosting and XGBoost). The compositional variables of the composites namely the epoxy content, jute fiber content (fixed at 20 wt.%), and LD sludge content (varied between 0 and 25 wt.%) were the input parameters for all the models. The experimental mechanical properties were the target outputs. Standard regression measures were used to test the model performance, and they included R 2 (Coefficient of Determination), MSE (Mean Squared Error), RMSE (Root Mean Squared Error) and MAE (Mean Absolute Error). All the machine learning models, including Decision Tree, Random Forest, Gradient Boosting and XGBoost, were implemented and trained using Google Colab, a cloud computing hosted system, which offered a convenient, scalable, and GPU-intensive platform to develop and test the regression models. Python was used as the main programming language and necessary libraries were imported to the Colab notebooks: scikit-learn, XGBoost, pandas, and Matplotlib were used to perform data preprocessing, model training, hyperparameter tuning, performance evaluation, and result visualization. Results and Discussion Tensile, impact, and flexural test results The mechanical properties of EP_JF_LDS composites illustrated in Figure 1 show a clear improvement with progressive modification of the composite from S0 to S4, followed by a marginal decline at S5. The tensile strength increases from 48 MPa for the S0 specimen to a value of 56.52 MPa for S1 specimen. This improvement indicates effective stress transfer between epoxy and LDS/jute fiber . The tensile strength further increases to 58.28, 60.37, and 61.84 MPa for S2, S3, and S4 composites, respectively. This gradual rise in tensile strength from S1 to S4 suggests improved interfacial adhesion and more uniform dispersion of the LDS particles, which together enhance the load-bearing capability under tensile loading. The slight reduction observed for the S5 specimen (60.62 MPa) can be attributed to possible agglomeration of reinforcement or matrix saturation effects, which tend to act as stress concentrators and limit further strength enhancement. A similar trend is observed with flexural strength. It steadily rises from 22.44 MPa (S0) to 31.81 MPa (S4) with higher resistance to bending stresses. The increased stiffness and enhanced bonding at the interface assist in the transfer of the load during flexural deformation. Again, the marginal decrease at S5 (30.85 MPa) indicates that excessive reinforcement content or uneven distribution can cause stress distribution, which causes formation of premature microcracks under bending loads. The same trend is observed with impact strength, which rises to its maximum of 18.026 kJ/m 2 at S4, starting from 13.352 kJ/m 2 at S0. This enhancement emphasizes the better energy absorption property of the composite due to better interfacial bonding and deflection of cracks. The small decrease at S5 (17.598 kJ/m 2 ) points to a loss of toughness which is probably due to particle agglomeration or loss of continuity in the matrix which limits plastic deformation and crack arrest during abrupt loading. In general, among all samples, sample S4 has the best composition, which provides the most appropriate balance of tensile and flexural and impact properties. The decline in properties beyond this point confirms that exceeding the optimum reinforcement level can adversely affect mechanical performance due to microstructural defects and inefficient stress transfer. Predictive performance for tensile strength Decision Tree regression model Decision Tree regression model satisfies ideal assessment values where R 2 = 1.0000, MSE = 0.0000, RMSE = 0.0000, and MAE = 0.0000 indicating excellent prediction precision in tensile strength estimation. In Figure 2(a), it is clear that 0% prediction error is achieved for all samples, which means that the model has not only learned the underlying patterns and relationships in the training data, but has also perfected them. The model can explain 100% of the tensile strength data, as shown by an R 2 value of 1.0000, and zero error measures (MSE, RMSE, and MAE) show that the values between the expected and actual values are accurate. This good fit is likely to indicate that the intricate non-linear correlations between the tensile strength output and the input parameters have most likely been captured, including factors related to material composition, such as the epoxy content, jute fiber characteristics, and sludge percentage, which have been adequately modelled by the Decision Tree. This is ideal behavior on the training data; however, it can raise concerns regarding possible overfitting, where the model has mastered the training data to an extreme level, considering noise and outliers, and in the process, the model can reduce it capacity to generalize on new data. The model is working on physically viable limits of polymer based composite reinforced with elements of natural fibers and industrial wastes as demonstrated by the tensile strength estimates of between 50 and 55 MPa, and this is equivalent to standard estimates of composite material. The model can generate 100% prediction accuracy and this is strong deterministic relationships among the input parameters and the mechanical properties. This may suggest that elements like compositional ratios, filler particle distribution, and fiber-matrix interface bonding exhibit predictable patterns that the Decision Tree method can accurately map. The tensile behavior of these hybrid composites is probably governed by critical threshold values and interaction effects between factors that the model has identified. The Decision Tree is a potentially useful tool for material design optimization and property prediction within the tested parameter space, as the performance metrics confirm that the chosen features contain enough information to fully describe the tensile strength variations seen in the experimental data. Random Forest regression model The evaluation metrics presented show (Figure 2(b)) that the Random Forest regression model has a high and reliable prediction performance for tensile strength estimation. The model demonstrates a high level of explanatory power with an R 2 of 0.9374. Together, the error metrics MSE of 1.3403, RMSE of 1.1577, and MAE of 0.6543 verify that the model's predictions closely match the measured data. The R 2 score of 0.9374 indicates that the input properties of the model account for about 93.74% of the variation in the specimens' tensile strength. This high value indicates that the model has been successful in locating and utilizing the dataset's underlying patterns and relationships. The magnitude of prediction errors is revealed by the error metrics. The overall prediction and actual tensile strength have an average difference of about 0.65 MPa which is shown by the MAE of 0.6543. No remarkable outliers with severe prediction errors, which is proved by the RMSE of 1.1577 MPa, which is more susceptible to larger errors and as such is moderately low. The visual confirmation of this excellent performance is seen in the almost clustering of the predicted values to the actual trend line in the chart. This is a characteristic of an optimized Random Forest model. A Random Forest is an ensemble algorithm, unlike a single Decision Tree, it creates a number of Decision Trees and aggregates their forecasts. The perfect-score Decision Tree example raised concerns about overfitting, which is obviously addressed by this method. The metrics shown by modeling the complicated, non-linear relationship between input parameters such as material composition, fiber properties, and processing conditions and the tensile strength output without simply remembering the data set indicate that the model has been able to generalize on the training data. In the case of composite materials particularly those made on the basis of polymeric matrices reinforced with natural fibers, the expected tensile strength values that lie between 50 MPa and 58 MPa are physically achievable. The ability of the model to accurately predict within this range proves that the selected features have sufficient details to describe the primary factors influencing the tensile behavior. Because it can accurately forecast mechanical properties based on compositional inputs, the Random Forest model is a very effective and dependable tool for the design and optimization of such composite materials. Gradient Boosting regression model For the tensile strength estimation, the Gradient Boosting regression model exhibits remarkable predictive ability, attaining nearly flawless assessment metrics. With comparably low error measures of MSE = 0.0000, RMSE = 0.0001, and MAE = 0.0001, the model achieves an R 2 value of 1.0000. Figure 2(c) shows that the model has learned the fundamental patterns and relationships in the training dataset with exceptionally high precision, leading to low prediction error in all samples. The model accounts for nearly all of the variation in the tensile strength data when the R 2 value is near 1.0. A very close alignment between the projected and actual values is confirmed by the minimal error metrics (MSE, RMSE, and MAE), with the MAE showing an average prediction error of less than 0.0001 MPa. Such a good fit implies that the Gradient Boosting algorithm has been able to model the complex, non-linear relationships and sensitive interaction effects between the tensile strength output and input variables, such as the percentage of sludge filler, epoxy resin content, and jute fiber properties. The high prediction accuracy of the model is among the key features of the Gradient Boosting framework. Gradient Boosting builds trees one after another, each tree is directly trained on the residual (or errors) of the existing ensemble as opposed to bagging algorithms such as Random Forest, which build trees in parallel. This iterative, so-called boosting process may focus on the most difficult cases, resulting in a powerful prediction model that often outperforms traditional ensemble methods. According to the chart, the tensile strength forecasts seem to fall between 48 MPa and 61 MPa, which is in good agreement with usual values for composite materials. For polymer-based composites reinforced with natural fibers and industrial waste components, this shows that the model is functioning within physically reasonable constraints. The performance metrics confirm that the chosen features provide enough information for the model to accurately characterize the differences in tensile strength. The inherent regularization in contemporary Gradient Boosting implementations makes it a highly dependable and useful tool for material design optimization and property prediction within the tested parameter space, even though such high performance needs to be verified on a different test set to definitively rule out overfitting. XGBoost (Extreme Gradient Boosting) regression model The XGBoost (Extreme Gradient Boosting) regression model achieves outstanding assessment metrics and shows state-of-the-art predictive performance for tensile strength prediction (Figure 2(d)). With comparably low error measures of MSE = 0.0000, RMSE = 0.0008, and MAE = 0.0005, the model achieves an R² value of 1.0000. This shows that the model has nearly zero prediction error across all samples, indicating that it has learned the fundamental patterns and relationships within the training dataset with exceptional precision. Nearly all of the variance in the tensile strength data can be explained by the model when the R 2 value is close to 1.0000. A nearly perfect alignment between the projected and actual values is confirmed by the incredibly low error metrics (MSE, RMSE, and MAE), with the MAE showing an average prediction error of just 0.0005 MPa. This better fit indicates that the XGBoost algorithm has successfully mapped the interaction between input parameters such as epoxy resin content, jute fiber properties, and sludge filler percentage and the tensile strength output in addition to capturing the intricate, non-linear relationships. One distinguishing feature of the XGBoost architecture is the model's capacity to reach such high prediction accuracy. According to Figure 2(d), the tensile strength predictions seem to fall between 48 and 61 MPa, which is in good agreement with accepted values for composite materials. This suggests that for polymer-based composites reinforced with natural fibers and components from industrial waste, the model's predictions are limited within physically reasonable limits. The performance measurements confirm that the chosen features offer a comprehensive and adequate explanation for the observed variances in tensile strength. XGBoost's inherent regularisation and cross-validation capabilities make it an especially reliable and useful tool for the inverse design and optimization of new composite materials, enabling highly accurate prediction of mechanical properties from compositional inputs, even though validation on a hold-out test set is still a crucial step. Comparative performance analysis The MSE, RMSE, and MAE values of the four machine learning models Decision Tree, Random Forest, Gradient Boost, and XGBoost that are used to predict the tensile strength of composite materials show (Figure 2(e)) a distinct hierarchy in predictive capability. Although the Decision Tree model provides a simplistic predictive model, the error measures are the highest, implying that this model has a low capacity to extrapolate the complex non-linear relationships between material composition factors such as epoxy content, jute fiber properties, and sludge percentage, and the tensile strength output. This is likely due to the fact that the Decision Tree model is more inclined to overfitting, and the Random Forest model, in contrast, exhibits a significant improvement in performance due to its bagging effect, where multiple trees are constructed simultaneously to reduce variance and decrease overfitting, to achieve much lower error measures and more affordable predictor model. Even more gains in performance are obtained by the Gradient Boost algorithm, which is a sequence of boosting methods in which a new tree corrects the errors of its predecessors. This iterative refinement enables it to capture complex patterns and subtle interaction effects more effectively, resulting in superior accuracy and the lowest error metrics among the conventional models. XGBoost is the most accurate and dependable tool for the inverse design and optimization of these composite materials because of its gradual advancement from a single tree to complex ensemble methods, which confirms that the input features contain enough information to model the tensile behavior. Predictive performance for flexural strength Decision Tree regression model The Decision Tree regression model achieves flawless evaluation metrics with an R 2 of 1.0000, MSE of 0.0000, RMSE of 0.0000, and MAE of 0.0000, demonstrating outstanding prediction accuracy for flexural strength estimation. This Figure 3(a) shows that all specimens, from S0 to S5, have zero prediction error since the model has flawlessly learned and duplicated the underlying patterns in the training dataset. The model explains 100% of the variance in the flexural strength values, as indicated by the R 2 value of 1.0000, and the zero error metrics verify a precise, point-for-point alignment between the anticipated and actual values. This perfect fit indicates that the Decision Tree algorithm has successfully captured the deterministic correlations between the flexural strength output and the input parameters, which are probably processing factors and material composition. Such perfect performance on the training data, however, strongly suggests overfitting, in which the model has probably learned the training dataset too precisely, including its noise and experimental changes, rather than the generalizable underlying principles. As a result, even though the model functions within physically reasonable constraints for the flexural strength of composites, its usefulness is constrained until it can be verified on an independent, unobserved test set to verify its capacity to generalize beyond the precise data on which it was trained. Random Forest regression model Strong assessment metrics, such as a R² of 0.9292, an MSE of 0.7060, an RMSE of 0.8403, and an MAE of 0.4687, show that the Random Forest regression model has strong and extremely effective predictive performance for evaluating the flexural strength of composite specimens as shown in Figure 3(b). The model successfully explains over 93% of the variance in the flexural strength data, as indicated by the high R 2 value. This shows that the input parameters contain highly useful features for predicting this mechanical attribute. The model's accuracy is further supported by the low error metrics; the RMSE of 0.8403 indicates that large errors are rare, while the MAE of 0.4687 MPa indicates that, on average, the model's predictions differ from the actual measured values by less than half a megapascal. This result is typical of the Random Forest's ensemble structure, which improves generalization on unseen data by combining predictions from several decorrelated decision trees to reduce the danger of overfitting, a common problem with single Decision Trees. The model has successfully learned the underlying non-linear relationships and interaction effects within the material system, as demonstrated by the close alignment between the actual and predicted values across specimens S1 through S5, which is clearly visible in the chart. This makes the model a dependable and effective tool for the design and optimization of composite materials. Gradient Boosting regression model With an ideal R² value of 1.0000 and nearly zero error metrics of MSE = 0.0000, RMSE = 0.0001, and MAE = 0.0001, the Gradient Boosting regression model exhibits theoretically flawless predictive accuracy for evaluating the flexural strength of composite specimens. This is shown in (Figure 3(c)) where the model has almost perfectly captured the underlying patterns and connections in the training dataset, producing predictions that are nearly identical to the actual observed values for every specimen from S0 to S5. The model describes 100% of the variation in the flexural strength data based on the value of an R 2 value of 1.0000 and the marginally small values used to measure the error metrics to almost a perfect congruence of the expected and actual values. Gradient Boosting is characterized by a remarkable fit, and this is due to the fact that it is based on the creation of decision trees in succession. Every new model is uniquely trained to fit the residual errors of the model before it and allows a more complex, non-linear relationship to be represented with an ensemble as well as subtle interaction effects among input parameters being captured with extreme accuracy. The model may have memorized the training data, its noise, and its specific variations in experiments to the extent that its performance on new, unknown data may be compromised. Nevertheless, such a perfect match to the training data is a cause of great concern with regard to overfitting. Hence, prior to the consistent application of this model to the inverse design of new composite materials, its practical applicability and generalizability need to be stringently assessed against a separate test set, even though it is a powerful means of analysis within the parameters space being tested. XGBoost regression model Having an ideal value of R 2 (1.0000) and a low error value of MSE = 0.0000, RMSE = 0.0008, and MAE = 0.0005, the XGBoost regression model presents impressive results in its almost perfect predictive accuracy in the assessment of the flexural strength of composite specimens. Figure 3(d) demonstrates that the model has significantly succeeded in capturing the underlying trends in the training data, and the predictions it makes are almost precise values of its measurements of all the specimens. The very small error values, particularly, the MAE of 0.0005 MPa, demonstrate that the mean error of prediction is not significant, which means that the model is highly accurate. The R 2 value of 1.0000 confirms that the model covers 100% of the variance in data of the flexural strength. This demonstration shows that the XGBoost algorithm has sophisticated functionality in building the relationships and slight interaction effects that are complex and non-linear between the input parameters, and is capable of doing so with higher effectiveness due to integrated L 1 and L 2 regularization to prevent overfitting, a more efficient process of tree-building, and a more advanced treatment of missing data. The regularization mechanisms inherent to the model ensure that it is a highly trustworthy and valuable model in the prediction and optimization of mechanical properties in composite material design although the near-perfect training result needs to be validated on an independent test set to ensure conclusively that the model is generalized. Comparative performance analysis The error measures of the four machine learning models provided in Figure 3(e) clearly demonstrate the existence of a strong hierarchy in predicting mechanical properties. The Decision Tree model was found to have the most errors; approximately 0.7 in MSE, 0.84 in RMSE, and 0.47 in MAE, which implies that it has a low predictive ability in generalization and shows a strong overfitting due to simple, single-tree design. The Random Forest model is much more effective, and the error indicators are reduced nearly by half (MSE is approximately 0.35, RMSE is approximately 0.59, and MAE is approximately 0.25). This demonstrates that its ensemble bagging strategy is a good method for reducing variance and enhancing predictive strength. The Gradient Boost model further enhances this accuracy by employing a sequential boosting mechanism, which progressively corrects residual errors to capture more complex patterns in the data resulting in even lower errors of approximately 0.15 for MSE, 0.39 for RMSE and 0.18 for MAE. Ultimately, the XGBoost model has the smallest error measures with MSE approximately 0.05, RMSE approximately 0.22, and MAE approximately 0.08. This is a confirmation of its advanced algorithmic enhancements, including incorporated L1 and L2 regularization and improved learning procedure, which together provide the model with the best and most reliable predictions. XGBoost is the most accurate model for mechanical property prediction in this composite material system, and this gradual decrease in error which will reach its peak in performance by this model represents a significant advantage of advanced ensemble methods. Predictive performance for impact strength Decision Tree regression model Considering an ideal R² value of 1.0000 and zero error metrics (MSE = 0.0000, RMSE = 0.0000, MAE = 0.0000), the Decision Tree regression model shows theoretically perfect predictive accuracy for forecasting the mechanical characteristics of the composite specimens. This Figure 4(a) shows that the model has learned the training dataset with perfect accuracy, generating predictions for each of the six specimens (S0 through S5) that precisely match the real values. Based on the input parameters, which are the particular material compositions of the specimens, the model explains 100% of the variance in the output characteristic, as indicated by the R 2 value of 1.0000. This perfect fit overwhelmingly suggests serious overfitting even if it implies the Decision Tree has properly mapped the deterministic relationship between the compositional inputs and the mechanical output. Rather of learning a generalizable rule, the model has probably committed all of the training data—including any experimental noise—to memory.As a result, even while the model performs flawlessly on this particular set of data, its practical applicability is severely constrained because it would most likely be unable to accurately forecast new, unexplored specimen compositions, which would compromise its dependability for material design and optimization. Random Forest regression model Strong performance measures, such as an R 2 of 0.9517, an MSE of 0.1281, an RMSE of 0.3579, and an MAE of 0.2638 kJ/m², show that the Random Forest regression model has exceptional predictive power for forecasting the impact strength of composite specimens. The high R 2 value shows (Figure 4(b)) that the model accounts for about 95.17% of the variance in the impact strength data, indicating that the fundamental factors controlling this mechanical property are successfully captured by the input features, which are probably related to material composition and specimen structure. Random Forest's ensemble approach, which builds several decision trees via bootstrap aggregation (bagging) and averages their predictions, effectively reduces the overfitting that is frequently connected to individual trees, improving the robustness and applicability of the model. Given the common range of impact strength for composite materials, the model's predictions differ from the actual impact strength values by just about 0.26 kJ/m² on average, according to the low error metrics, especially the MAE of 0.2638 kJ/m². This finding indicates the ability of the model to capture non-linear and complex relationships and interaction effects in the data which provides it one of the most reliable models in optimizing material formulation and predicting the mechanical behavior of composite systems. .Gradient Boosting regression model The Gradient Boosting regression model demonstrates an ideal fit to predict the impact strength of the composite specimens with a R 2 = 1.0000 and with zero error values (MSE = 0.0000, RMSE = 0.0000, MAE = 0.0000 kJ/m 2 ). The Figure 4(c) shows that the model's predictions for specimens S0 through S5 had zero average prediction error, no unexplained variance, and perfect, point-for-point alignment with the actual measured impact strength values. The Gradient Boosting algorithm's sequential learning process, in which each new decision tree in the ensemble is specially built to correct the residual errors of the combined previous trees, is characterized by this faultless performance, which enables it to gradually model the intricate, non-linear relationship between the input parameters and impact strength with ultimate precision on the training data. Instead of learning a generalizable function that would accurately predict the impact strength of novel, untested material compositions, such a perfect result is a classic sign of extreme overfitting, implying that the model has effectively memorized the training dataset including any experimental noise or particular peculiarities of the six specimens. This makes the model an ideal analytical tool for the available data, but without validation on a different test set to verify its capacity to generalize beyond the specific specimens it was trained on, its practical utility for predictive material design is seriously problematic. XGBoost regression model Using an optimal R 2 value of 1.0000, the XGBoost model exhibits nearly perfect prediction accuracy for estimating impact strength, confirming that the model explains 100% of the variance in the target property based on the input data as shown in Figure 4(d). With an RMSE of 0.0008 and MSE and MAE of a minimal 0.0000 and 0.0006 kJ/m², respectively, the error metrics show that the predictions are almost exactly the same as the actual measured values across all specimens from S0 to S5, with an average error of only 0.0006 kJ/m². XGBoost's sequential, additive tree-building method, which systematically eliminates residuals from earlier models while controlling for overfitting through built-in L1 and L2 regularization, and its sophisticated regularization algorithms are responsible for its remarkable performance. Their final result is an exceptionally complex model which does not overfit as is common with ideal training scores and represents complex, non-linear relationships and nuanced interaction effects in the data of material composition. As a result, XGBoost is a very dependable tool for the precise prediction and optimization of impact strength in novel composite material formulations since it not only reaches theoretical perfection on the training set but also shows a great potential for robust generalization. Comparative performance analysis A clear performance hierarchy for forecasting the mechanical properties of composite specimens is revealed by comparing the error metrics of the four machine learning models as shown in Figure 4(e). Due to a high tendency for overfitting from its single-tree structure, the Decision Tree model has the highest error values. Its metrics, which are probably near the top of the scale at 0.30 for MSE, 0.55 for RMSE, and 0.25 for MAE, highlight its fundamental limitation in generalizing beyond the training data. The Random Forest model, which uses ensemble bagging, reduces these errors by roughly half, to about 0.15 in MSE, 0.39 in RMSE, and 0.12 in MAE. This is the confirmation that the model is stronger and better in ability to define the underlying data patterns using the aspect of variance reduction. The Gradient Boost model builds on this by applying a sequential boosting algorithm to the error again to get it down to approximately 0.07 on MSE, 0.26 on RMSE and 0.08 on MAE. This demonstrates that it is a highly advanced method of modeling complex non-linear interactions through the correction of prediction residuals during a sequence of steps. XGBoost model eventually attains the lowest error rates, the scores go down to nearly zero- 0.02 in MSE, 0.14 in RMSE and 0.03 in MAE. It is effective because of its very efficient learning algorithm and advanced regularization which in combination enable the model to possess unmatched predictive precision and opens up to its status as the optimal model to possess high-reliability predictability of mechanical properties in composite material systems. Overall feature importance Heatmap The feature importance analysis provides clear-cut mechanistic roles of each composite constituent in the control of mechanical performance (Figure 5(a)). The content of epoxy shows the highest effect on all properties, especially in terms of impact strength, and the value of importance of 0.746. This large dominance demonstrates the important role of the matrix in the energy absorption and crack resistance, in which continuous and tough epoxy phase improves the ductile and the stress distribution of the interfaces, making the impact resistance increase dramatically between 13.352 kJ/m 2 in pure epoxy to 18.026 kJ/m 2 in the composite of the optimum. In tensile and flexural strength, epoxy is the most affecting variable (0.480 and 0.471, respectively), which is expected given that this is the primary load-bearing medium, which upholds the tensile strength to rise to 61.84 MPa and flexural strength to 31.81 MPa in S4 composite. Jute fiber demonstrates high significance on tensile (0.341) and flexural strength (0.370), which can be attributed to the fact that the fiber reinforces the matrix through fiber bridging and load transfer, and the fiber indicates a low contribution towards impact strength (0.041) meaning that during high-rate loading, the bonding between the fiber and the matrix can be damaged prematurely, which restricts the energy absorption. Although Sludge content has the smallest effect on strength properties (0.179 tensile and 0.159 flexural), it has a significant effect on impact strength (0.213), meaning that the hard sludge particles contribute to crack deflection and localized plastic deformation, though overloading the material above 20 wt.% reduces interfacial adhesion and boosts brittleness as seen by the deterioration of mechanical properties at 25 wt. % sludge. Correlation matrix The correlation matrix shown in Figure 5(b) indicates that negativities of composition are strong as the correlations between epoxy, jute, and sludge are high with negative correlation of -0.92, -0.90; this can be attributed to the fact that an increase in the percent composition causes a reduction in the others as dictated by the formulation constraint: the total weight percentage has to be 100. This inherent trade-off is a direct effect on behavior of mechanical properties. All of the three mechanical properties, tensile strength (-0.98), flexural strength (-0.96), and impact strength (-0.96) have very strong negative relationships with sludge content, which confirms that above the optimal sludge loading (20 wt.%) the mechanical properties are reduced by agglomeration of particles, lack of adhesion, and the increase in the content of voids. On the other hand, tensile (0.98), flexural (0.96), and impact strength (0.96) exhibit very strong positive correlations with epoxy content, which highlights the importance of polymer matrix in the maintenance of structural integrity, transfer of stress, and toughness provided by good interfaces with both jute fibers and sludge particles. Another good correlation can be found between jute fiber content and tensile (0.92), flexural strength (0.95) which indicates that jute fiber can reinforce the material through the mechanism of fiber-matrix interaction and load bearing, but the correlation between jute fiber content and impact strength is moderate (0.88), indicating that the role of jute fiber is not as dominant as that of the epoxy matrix under dynamic loading. These correlation trends indicate that there is an undeniable mechanistic trade-off that the more epoxy used, the better the mechanical strength achieved, but the more sludge used, though it is good in hardness and density, the more it negatively affects the strength properties beyond a certain optimal level, and this is the way formulation strategies can be pursued towards balanced performance in sustainable hybrid composites. Optimization of composite composition The composite formula (S4) was optimized to reach high mechanical characteristics in comparison to the neat epoxy base, and the improved mechanical properties of the composite formula can be attributed to the synergistic reinforcement processes of the jute fiber and LD sludge (Figure 5(c)). The flexural strength displayed maximum improvement of 41.8 which showed an increase of 22.44 MPa to 31.81 MPa under bending due to high transfer of the load and bridging effect of the jute fibers which in turn reduced the concentration of stress and also delayed the formation of a crack, which is attributed to enhanced interfacial adhesion due to good dispersion of the sludge particles. It has increased tensile strength of 28.8% with the sludge increasing tensile strength of 48 MPa to 61.84 MPa; this is mainly because of the reinforcing nature of jute fibers, which bear the axial loads and the ability of sludge to fill up the micro voids, which in combination with both increase the continuity and resist tensile deformation of the matrix. Impact strength was enhanced by 35.0% to 18.026 kJ/m 2 because of energy dissipation mechanisms (fiber pull-out, crack deflection around sludge particles, and matrix micro-cracking) leading to a significant increase in toughness. These percentage improvements represent the positive incorporation of natural fiber and industrial waste filler, with the jute being the main form of reinforcement, the sludge modifying microstructure and interfacial characteristics, and the epoxy matrix being the type of reinforcement that bonds the hybrid composite system, with the resulting tensile, flexural, and impact properties to be balanced. Conclusions The current research demonstrated successful fabrication of high performance and eco-friendly epoxy composites using jute fiber and LD sludge as hybrid reinforcements. The experimental findings indicate that the S4 formulation (20 wt.% jute, 20 wt.% sludge) provides the best mechanical performance, with tensile strength increased from 48 MPa to 61.84 MPa (28.8%), flexural strength increased from 22.44 MPa to 31.81 MPa (41.8%), and impact strength increased from 13.352 kJ/m 2 to 18.026 kJ/m 2 (35%). It is suggested that these gains are due to synergistic reinforcement mechanisms such as good stress transfer, better fiber-matrix adhesion and deflection of cracks by sludge particles. At the maximum filler loading (25 wt.%), mechanical properties are reduced, and it is essential to have a balanced incorporation of fillers. The application of machine learning models, specifically XGBoost, yielded a highly precise predictive model (R² = 1.000, MAE = -0.0005 MPa) for predicting mechanical behavior using compositional data, thereby facilitating effective material optimization. The study confirms a dual waste-valorisation and data-driven design approach, which is part of the formation of sustainable composites for structural and industrial use. This study demonstrates that industrial waste-filled EP-JF composites can be utilized as low-cost materials, contributing to sustainable material development. The optimised composite system exhibits enhanced mechanical properties, making it a suitable candidate for load-bearing and semi-structural applications such as interior automotive panels, lightweight housings, brackets, enclosures, partition boards, furniture components, building panels, and other structural components where moderate strength, stiffness, and durability are required. Limitations of this work This study is limited by the narrow compositional range investigated and the relatively small dataset used for the machine learning model, which may restrict broader generalization. The work focused mainly on mechanical properties, without a detailed evaluation of long-term durability, thermal behavior, or large-scale manufacturing feasibility. Further studies with expanded compositions, larger datasets, and extended performance testing are needed to strengthen practical applicability Declarations Ethics approval and consent to participate Not Applicable. Patient consent statement Not Applicable. Consent to publish Not applicable. Funding statement This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Whereas Open access funding will be provided by Manipal Academy of Higher Education, Manipal Author Contribution Abhilash Purohit contributed to the methodology, wrote the original draft, conducted experiments, and performed data analysis. S. Sathees Kuma contributed to writing the original draft, experiments, and data analysis. Pravat Ranjan Pati contributed to conceptualization, writing, and data curation. Arvind Kumar contributed to conceptualization and provided funding for the experiment. Suresh Palanimuthu contributed to supervising the experiment and provided funding for the experiment. Byomakesh Mahapatra contributed to validation, reviewing, and editing Acknowledgement The article is supported by MAHE OA funding Data Availability The data used and/or analyzed during the current study are available from the corresponding author upon reasonable request. References Mohanty AK, Misra M, Drzal LT. Sustainable bio-composites from renewable resources: opportunities and challenges in the green materials world. J Polym Environ 2002; 10: 19–26. Ochi D, Barbieri D, Reis AF, et al. Chapter 1 - Agro-industrial waste as fillers for green composites. In: Inamuddin, Altalhi T, Alrooqi A (eds) Green Sustainable Process for Chemical and Environmental Engineering and Science . Elsevier, pp. 1–26. Pugar D, Ćurković L, Gabelica I, et al. Potential Use of Jarosite Industrial Waste in Developing Hybrid Composites. Sustain ; 16. Ramesh M, Palanikumar K, Reddy KH. Plant fibre based bio-composites: Sustainable and renewable green materials. Renew Sustain Energy Rev 2017; 79: 558–584. Sabeel Ahmed K, Khalid SS, Mallinatha V, et al. Dry sliding wear behavior of SiC/Al2O3 filled jute/epoxy composites. Mater Des 2012; 36: 306–315. Goriparthi BK, Suman KNS, Mohan Rao N. Effect of fiber surface treatments on mechanical and abrasive wear performance of polylactide/jute composites. Compos Part A Appl Sci Manuf 2012; 43: 1800–1808. Das B, Prakash S, Reddy PSR, et al. An overview of utilization of slag and sludge from steel industries. Resour Conserv Recycl 2007; 50: 40–57. Reghunadhan A, Datta J, Jaroszewski M. Polyurethane glycolysate from industrial waste recycling to develop low dielectric constant , thermally stable materials suitable for the electronics. Arab J Chem 2020; 13: 2110–2120. Vigneshwaran S, Uthayakumar M, Arumugaprabu V. Potential use of industrial waste-red mud in developing hybrid composites: A waste management approach. J Clean Prod 2020; 276: 124278. Vinod A, Tengsuthiwat J, Gowda Y, et al. Jute/Hemp bio-epoxy hybrid bio-composites: Influence of stacking sequence on adhesion of fiber-matrix. Int J Adhes Adhes 2022; 113: 103050. Tian J, Li C, Xian G. Reciprocating friction and wear performances of nanometer sized-TiO2 filled epoxy composites. Polym Compos 2021; 42: 2061–2072. Hsissou R, Seghiri R, Benzekri Z, et al. Polymer composite materials: A comprehensive review. Compos Struct 2021; 262: 113640. Jin F-L, Li X, Park S-J. Synthesis and application of epoxy resins: A review. J Ind Eng Chem 2015; 29: 1–11. Sarikaya E, Çallioğlu H, Demirel H. Production of epoxy composites reinforced by different natural fibers and their mechanical properties. Compos Part B Eng 2019; 167: 461–466. Sanjay MR, Madhu P, Jawaid M, et al. Characterization and properties of natural fiber polymer composites: A comprehensive review. J Clean Prod 2018; 172: 566–581. Kumar S, Saha A. Graphene nanoplatelets/organic wood dust hybrid composites: physical, mechanical and thermal characterization. Iran Polym J 2021; 30: 935–951. Vengadesan E, Arunkumar T, Muralidharan S, et al. Hybrid bio-composites reinforced with natural wood saw dust and eco-friendly graphite: evaluation of physical, mechanical, and thermal properties. Fibers Polym 2025; 26: 833–854. Nurazzi NM, Harussani MM, Aisyah HA, et al. Treatments of natural fiber as reinforcement in polymer composites-a short review. Funct Compos Struct 2021; 3: 24002. Gupta MK, Srivastava RK. Mechanical Properties of Hybrid Fibers-Reinforced Polymer Composite: A Review. Polym - Plast Technol Eng 2016; 55: 626–642. Thakur VK, Kessler MR. Self-healing polymer nanocomposite materials: A review. Polymer (Guildf) 2015; 69: 369–383. Purohit A, Satapathy A. Development and characterization of epoxy-based composites filled with Linz--Donawitz sludge. J Compos Mater 2017; 51: 899–911. Purohit A, Satapathy A. Processing, characterization, and parametric analysis of erosion behavior of epoxy-LD sludge composites using T aguchi technique and response surface method. Polym Compos 2018; 39: E2283--E2297. Purohit A, Satapathy A. Mechanical and wear characteristics of epoxy composites filled with industrial wastes: a comparative study. In: IOP Conference Series: Materials Science and Engineering . 2017, p. 12019. Purohit A, Swain PTR, Patnaik PK. Mechanical and sliding wear characterization of LD sludge filled hybrid composites. Mater Today Proc 2020; 26: 1654–1659. Purohit A, Tripathy V, Mishra SK, et al. Mechanical and Tribo-performance Analysis of Linz Donawitz Sludge-Filled Glass–Epoxy Composites using Taguchi Experimental Design. J Inst Eng Ser E 2022; 35: 1–8. Kumar SS, Shyamala P, Pati PR. Machine learning algorithms to predict the tensile strength of novel composite materials. Next Mater 2025; 9: 101105. Beldar P, Kadbhane S. Utilizing machine learning to forecast mechanical characteristics of NaOH-Treated jute fiber reinforced composite materials. Mater Lett 2024; 377: 137411. Milad A, Hussein SH, Khekan AR, et al. Development of ensemble machine learning approaches for designing fiber-reinforced polymer composite strain prediction model. Eng Comput 2022; 38: 3625–3637. Youssefi M, Safaie B. The Study on the Mechanical Properties of Multi-walled Carbon Nanotube/Polypropylene Fibers. J Inst Eng Ser E 2018; 99: 37–42. Rahman ANMM, Ahmed F, Alimuzzaman S, et al. Enhancement of Thermo-mechanical Properties of Okra Fiber by Photografting Technique. J Inst Eng Ser E 2021; 102: 45–59. Mahapatra SS, Patnaik A. Study on mechanical and erosion wear behavior of hybrid composites using Taguchi experimental design. Mater \& Des 2009; 30: 2791–2801. Erdo\ugdu YE, Korkmaz EE, Temiz \cSemsettin. Effect of graphene nanoplatelet filling on mechanical properties of natural fiber reinforced polymer composites. Mater Test 2021; 63: 322–328. Mirzamohammadi S, Eslami-Farsani R, Ebrahimnezhad-Khaljiri H. The experimental assessment of carbon nanotubes incorporation on the tensile and impact properties of fiber metal laminate fabricated by jute/basalt fabrics-aluminum layers: as hybrid structures. Proc Inst Mech Eng Part C J Mech Eng Sci 2023; 237: 1877–1886. Olhan S, Antil B, Behera BK. Low-velocity impact and quasi-static post-impact compression analysis of woven structural composites for automotive: Influence of fibre types and architectural structures. Compos Struct 2025; 352: 118676. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 30 Mar, 2026 Reviews received at journal 29 Mar, 2026 Reviews received at journal 29 Mar, 2026 Reviews received at journal 22 Mar, 2026 Reviewers agreed at journal 20 Mar, 2026 Reviewers agreed at journal 20 Mar, 2026 Reviewers agreed at journal 20 Mar, 2026 Reviewers agreed at journal 01 Mar, 2026 Reviewers agreed at journal 01 Mar, 2026 Reviews received at journal 24 Feb, 2026 Reviewers agreed at journal 24 Feb, 2026 Reviewers agreed at journal 24 Feb, 2026 Reviewers invited by journal 24 Feb, 2026 Editor assigned by journal 24 Feb, 2026 Editor invited by journal 23 Feb, 2026 Submission checks completed at journal 20 Feb, 2026 First submitted to journal 20 Feb, 2026 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-8903744","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":596538866,"identity":"1c334e4d-4fe0-40ed-a0ec-bbda3be5e0fc","order_by":0,"name":"Abhilash Purohit","email":"","orcid":"","institution":"Graphic Era (Deemed to be University)","correspondingAuthor":false,"prefix":"","firstName":"Abhilash","middleName":"","lastName":"Purohit","suffix":""},{"id":596538873,"identity":"712bbcab-2407-40fe-9960-3313870fcee2","order_by":1,"name":"S. Sathees Kumar","email":"","orcid":"","institution":"Saveetha Engineering College","correspondingAuthor":false,"prefix":"","firstName":"S.","middleName":"Sathees","lastName":"Kumar","suffix":""},{"id":596538876,"identity":"8858c133-8e4a-4708-891b-af7c7140992e","order_by":2,"name":"Pravat Ranjan Pati","email":"","orcid":"","institution":"Graphic Era (Deemed to be University)","correspondingAuthor":false,"prefix":"","firstName":"Pravat","middleName":"Ranjan","lastName":"Pati","suffix":""},{"id":596538877,"identity":"fd7315f2-474f-4d82-ab91-3fb04028877e","order_by":3,"name":"Arvind Kumar","email":"","orcid":"","institution":"CGC University","correspondingAuthor":false,"prefix":"","firstName":"Arvind","middleName":"","lastName":"Kumar","suffix":""},{"id":596538878,"identity":"2a97d9c6-b549-4dac-b5b0-3c7f69b40ffe","order_by":4,"name":"Suresh Palanimuthu","email":"","orcid":"","institution":"J J College of Engineering and Technology","correspondingAuthor":false,"prefix":"","firstName":"Suresh","middleName":"","lastName":"Palanimuthu","suffix":""},{"id":596538880,"identity":"8d4ecb3c-d185-497b-a184-c0505d748f03","order_by":5,"name":"Byomakesh Mahapatra","email":"data:image/png;base64,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","orcid":"","institution":"Manipal Academy of Higher Education","correspondingAuthor":true,"prefix":"","firstName":"Byomakesh","middleName":"","lastName":"Mahapatra","suffix":""}],"badges":[],"createdAt":"2026-02-17 18:53:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8903744/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8903744/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":103539839,"identity":"58d14c3c-f9d9-4fb9-b024-e0902ea164e3","added_by":"auto","created_at":"2026-02-26 19:42:23","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":91433,"visible":true,"origin":"","legend":"\u003cp\u003eMechanical properties of EP_JF_LDS composite\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8903744/v1/68b35512f7f421bf38f0e6aa.png"},{"id":104397749,"identity":"1da0a17a-84b7-44d1-91e8-873601b386c2","added_by":"auto","created_at":"2026-03-11 11:55:22","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":180491,"visible":true,"origin":"","legend":"\u003cp\u003eRegression models for tensile strengthprediction and performance comparison\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8903744/v1/0de5cf35805769e3db0d4a96.png"},{"id":103539840,"identity":"f8afa194-0202-4c99-978d-5c9742b9369a","added_by":"auto","created_at":"2026-02-26 19:42:24","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":138732,"visible":true,"origin":"","legend":"\u003cp\u003eRegression models for flexural strength prediction and performance comparison\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8903744/v1/0fd030c8802c1860ea8ee44e.png"},{"id":103539843,"identity":"25d3c794-57f0-40e0-b7c5-ac9b88c3e2d5","added_by":"auto","created_at":"2026-02-26 19:42:24","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":156492,"visible":true,"origin":"","legend":"\u003cp\u003eRegression models for impactstrength prediction and performance comparison\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8903744/v1/317c21edb3185ae80ea72648.png"},{"id":104397515,"identity":"335667d0-8550-4643-a182-3a0efaa58eaa","added_by":"auto","created_at":"2026-03-11 11:50:09","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":146223,"visible":true,"origin":"","legend":"\u003cp\u003e(a) Overall feature importance, (b) Correlation matrix, and (c) Optimization of composite composition\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8903744/v1/8e6c872cc1b309f279315e8d.png"},{"id":104407240,"identity":"8999c3c9-c205-4820-9664-046a5df4e9bd","added_by":"auto","created_at":"2026-03-11 12:36:05","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1585282,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8903744/v1/eb76d5d6-dbfd-4663-af86-998a549a8cc7.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning Approach for Mechanical Property Assessment of Industrial Waste-Filled Epoxy-Jute Composites","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe increased interest in sustainability and environmental responsibility in materials science has enhanced studies on eco-friendly composites that are reinforced with both natural fibers and industrial waste fillers [\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Jute fiber (JF) is one natural fiber that has been revealed to be a promising reinforcement material because it is biodegradable, has low density, high specific strength, and is also cost-effective and thus suitable as reinforcement in composites that use polymer as matrix [\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Simultaneously, the valorisation of industrial by-products like Linz-Donawitz (LD) sludge, which is an example of metallurgical waste product during the steel production, is a two-fold solution of disposing of waste and improving the composite quality [\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Epoxy resins are widely used as a matrix material in composites due to their high mechanical strength, chemical resistance, and adhesive ability [\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. However, epoxy is a brittle material, has low impact strength and medium stiffness, which limit its application in the construction sector [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. In order to overcome these shortcomings, fillers as well as the hybrid reinforcements have been extensively explored. Hybrid composites are those that have two or more different reinforcements, and they may have synergistic effects that enhance mechanical, thermal and physical performance [\u003cspan additionalcitationids=\"CR16 CR17\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The inclusion of LD sludge in the epoxy-jute composites is a novel method for developing sustainable hybrid materials. LD sludge may serve as a low-cost filler with improvements in hardness, density and mechanical strength due to dispersion strengthening and microstructural modification [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. However, the mechanical performance of such composites is highly dependent on factors such as filler loading, interfacial adhesion, particle dispersion, and matrix-filler compatibility [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e \u003cp\u003ePurohit and Satapathy have investigated the mechanical, sliding wear [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] and erosion wear [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] behavior of LD sludge filled epoxy composites [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], and found that the addition of only particulate filler reduces the mechanical properties as a result of stress concentration. Abhilash Purohit et al. [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] examined epoxy composites reinforced with wood apple dust (WAD) and LinzDonawitz (LD) sludge and studied their mechanical and wear properties. The composite containing 5 wt.% WAD and 12 wt.% LD sludge demonstrated higher flexural (24%), compressive (9.88%), and tensile (30.2) strengths, and RSM analysis proved the high dependence of wear strength on filler additio. Purohit et al. [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] examined the mechanical and dry sliding wear behavior of epoxy composites reinforced with LD sludge, BF slag and LD slag. A Taguchi design was used to measure mechanical properties and wear performance, where LD sludge-filled epoxy composites were known to have better mechanical and wear properties, and their strength increased up to 15% in the case of epoxy composites with 16 wt.% of sludge fillers. The application of machine learning (ML) in materials science has revolutionized the design and optimization of novel composites because it has the capability to predict, with high precision and based on data, the mechanical characteristics of a given material by using compositional variables.\u003c/p\u003e \u003cp\u003eRecent research shows that ensemble tree-based algorithms like Random Forest, Gradient Boosting, and XGBoost are better than traditional regression algorithms. Based on the study of Kumar et al. [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], the proposed study used machine learning to predict the publicly available natural fiber reinforced polymer (NFRP) datasets with the help of the following input parameters: epoxy content, density, and filler ratio. Random Forest was found to have the highest predictive power (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.92, MAE\u0026thinsp;=\u0026thinsp;1.64 MPa) compared to XGBoost, Gradient Boosting and Polynomial Regression and was tested using five-fold cross-validation. Beldar and Kadbhane [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] developed predictive models of NaOH-treated jute fiber-reinforced glass epoxy composites. XGBoost demonstrated the highest performance with an R\u003csup\u003e2\u003c/sup\u003e of 0.9714 (tensile), 0.9803 (hardness) and 0.9718 (flexural) with the MSE values of 209.84, 52.02 and 127.71, respectively, due to regularization and tree pruning using five-fold cross-validation. The study by Milad et al. [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] involved XGBoost, multivariate adaptive regression spline (MARS), and random forest (RF) models in predicting the strain enhancement ratio of FRP composites based on experimental data from the literature. This paper investigated and compared the XGBoost, MARS and FRP models to predict the strain enhancement ratio in FRP composites using 729 experimental datasets provided by the literature. There were five input combinations formed due to the main material, strain, strength, and confinement parameters. All the models performed well, with MARS achieving the best accuracy, whereas RF also performed well with only strain properties.\u003c/p\u003e \u003cp\u003eBased on these previous studies, it is noted that the wear resistance of the composites is enhanced when Linz Donawitz (LD) sludge because it is predominantly composed of iron oxides and silicon dioxide. The mechanical properties however, decrease with an increase in LD sludge percentage in the composite because of its irregular shape. This reducion in mechanical strength when a particulate filler is added can be avoided by adding another reinforcing phase in the form of a fiber [\u003cspan additionalcitationids=\"CR30\" citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. According to these observations, to eliminate the decline in mechanical properties, the LD sludge was mixed in different proportions and strengthened using jute/ epoxy composites. The tensile, flexural and impact strengths were used to examine the mechanical properties. Nevertheless, machine learning models were used to confirm the findings of the experimental results.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003ch2\u003eMaterials selection and fabrication method\u003c/h2\u003e\n\u003cp\u003eIn this study, LY-556 epoxy (EP) was used as the matrix material along with the corresponding hardener. The epoxy and hardener were mixed in a 10:1 ratio as recommended by the supplier (Scientific supplier, Bhubaneswar, India). A natural fiber namely jute fiber (JF) was chosen to improve the mechanical properties of the composites. However, an industrial waste called LD sludge (LDS) was used to improve the hardness of the composites. LDS particles were collected from the Rourkela steel plant, Odisha, India, and sieved using a standard mesh to obtain a uniform particle size around 100 \u0026micro;m. The LDS particles were kept in an oven at 103\u0026deg;C for four hours to remove moisture.\u003c/p\u003e\n\u003cp\u003eHybrid EP-JF-LDS composites were fabricated by the hand-lay-up technique. Six specimens were prepared with different percentages of matrix and reinforcement phases, and the details are given in Table 1. The neat epoxy specimens (S0) were prepared to compare the properties of the other composites (S1, S2, S3, S4, and S5) with the neat epoxy. The designation of the specimens, along with the percentages of LDS and JF, is shown in Table 1.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e Nomenclature and classification of composite specimens\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"408\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003eSpecimen Designation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 335px;\"\u003e\n \u003cp\u003eEP-JF-LDS Composition\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003eS0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 335px;\"\u003e\n \u003cp\u003e100 wt. % EP + 0 wt. % JF + 0 wt. % LDS\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003eS1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 335px;\"\u003e\n \u003cp\u003e75 wt. % EP + 20 wt. % JF + 5 wt. % LDS\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003eS2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 335px;\"\u003e\n \u003cp\u003e70 wt. % EP + 20 wt. % JF + 10 wt. % LDS\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003eS3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 335px;\"\u003e\n \u003cp\u003e65 wt. % EP + 20 wt. % JF + 15 wt. % LDS\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003eS4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 335px;\"\u003e\n \u003cp\u003e60 wt. % EP + 20 wt. % JF + 20 wt. % LDS\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003eS5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 335px;\"\u003e\n \u003cp\u003e55 wt. % EP + 20 wt. % JF + 25 wt. % LDS\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003ch2\u003eTesting methodology and characterization of composite samples\u003c/h2\u003e\n\u003cp\u003eThe tensile properties of the EP-JF-LDS hybrid composites were determined using the ASTM D3039 standard. Samples with 165 mm \u0026times; 19 mm \u0026times; 3.2 mm dimensions were prepared for the tensile tests. The accuracy of the test was within the range of \u0026plusmn; 2%. Flexural strength of the EP-JF-LDS composites was determined as per the ASTM D790 [32] standard. Specimens with a gauge length of 51 mm and dimensions of 127 mm \u0026times; 12.7 mm \u0026times; 3.2 mm were prepared to carry out the flexural tests. The impact strength was measured in accordance with ASTM D256 using a pendulum impact tester. Samples with a 45\u0026deg; notch and 64 mm \u0026times; 12.7 mm \u0026times; 3.2 mm dimensions were prepared for the test [33, 34]. Three samples of the same composition were used for tensile, impact and flexural tests, and the average of the three readings was used to determine the final tensile strength value.\u003c/p\u003e\n\u003ch2\u003eML models for mechanical properties estimation\u003c/h2\u003e\n\u003cp\u003eThis study used four different machine learning regression models to estimate the mechanical properties, namely tensile, flexural and impact strengths, of epoxy-jute-LD sludge composites. The models chosen are Decision Tree Regression, Random Forest Regression, Gradient Boosting Regression and XGBoost (Extreme Gradient Boosting) Regression. These models have been selected to compare and evaluate their predictive power and generalization (A simple interpretable baseline (Decision Tree) to more complex ensemble approaches: Random Forest, Gradient Boosting and XGBoost). The compositional variables of the composites namely the epoxy content, jute fiber content (fixed at 20 wt.%), and LD sludge content (varied between 0 and 25 wt.%) were the input parameters for all the models. The experimental mechanical properties were the target outputs. Standard regression measures were used to test the model performance, and they included R\u003csup\u003e2\u003c/sup\u003e (Coefficient of Determination), MSE (Mean Squared Error), RMSE (Root Mean Squared Error) and MAE (Mean Absolute Error). All the machine learning models, including Decision Tree, Random Forest, Gradient Boosting and XGBoost, were implemented and trained using Google Colab, a cloud computing hosted system, which offered a convenient, scalable, and GPU-intensive platform to develop and test the regression models. Python was used as the main programming language and necessary libraries were imported to the Colab notebooks: scikit-learn, XGBoost, pandas, and Matplotlib were used to perform data preprocessing, model training, hyperparameter tuning, performance evaluation, and result visualization.\u003c/p\u003e"},{"header":"Results and Discussion","content":"\u003ch2\u003eTensile, impact, and flexural test results\u003c/h2\u003e\n\u003cp\u003eThe mechanical properties of EP_JF_LDS composites illustrated in Figure 1 show a clear improvement with progressive modification of the composite from S0 to S4, followed by a marginal decline at S5. The tensile strength increases from 48 MPa for the S0 specimen to a value of 56.52 MPa for S1 specimen. This improvement indicates effective stress transfer between epoxy and LDS/jute fiber . The tensile strength further increases to 58.28, 60.37, and 61.84 MPa for S2, S3, and S4 composites, respectively. This gradual rise in tensile strength from S1 to S4 suggests improved interfacial adhesion and more uniform dispersion of the LDS particles, which together enhance the load-bearing capability under tensile loading. The slight reduction observed for the S5 specimen (60.62 MPa) can be attributed to possible agglomeration of reinforcement or matrix saturation effects, which tend to act as stress concentrators and limit further strength enhancement. A similar trend is observed with flexural strength. It steadily rises from 22.44 MPa (S0) to 31.81 MPa (S4) with higher resistance to bending stresses. The increased stiffness and enhanced bonding at the interface assist in the transfer of the load during flexural deformation. Again, the marginal decrease at S5 (30.85 MPa) indicates that excessive reinforcement content or uneven distribution can cause stress distribution, which causes formation of premature microcracks under bending loads. The same trend is observed with impact strength, which rises to its maximum of 18.026 kJ/m\u003csup\u003e2\u003c/sup\u003e at S4, starting from 13.352 kJ/m\u003csup\u003e2\u003c/sup\u003e at S0. This enhancement emphasizes the better energy absorption property of the composite due to better interfacial bonding and deflection of cracks. The small decrease at S5 (17.598 kJ/m\u003csup\u003e2\u003c/sup\u003e) points to a loss of toughness which is probably due to particle agglomeration or loss of continuity in the matrix which limits plastic deformation and crack arrest during abrupt loading. In general, among all samples, sample S4 has the best composition, which provides the most appropriate balance of tensile and flexural and impact properties. The decline in properties beyond this point confirms that exceeding the optimum reinforcement level can adversely affect mechanical performance due to microstructural defects and inefficient stress transfer.\u003c/p\u003e\n\u003ch2\u003ePredictive performance for tensile strength\u003c/h2\u003e\n\u003ch3\u003eDecision Tree regression model\u003c/h3\u003e\n\u003cp\u003eDecision Tree regression model satisfies ideal assessment values where R\u003csup\u003e2\u0026nbsp;\u003c/sup\u003e= 1.0000, MSE = 0.0000, RMSE = 0.0000, and MAE = 0.0000 indicating excellent prediction precision in tensile strength estimation.\u003c/p\u003e\n\u003cp\u003eIn Figure 2(a), it is clear that 0% prediction error is achieved for all samples, which means that the model has not only learned the underlying patterns and relationships in the training data, but has also perfected them. The model can explain 100% of the tensile strength data, as shown by an R\u003csup\u003e2\u003c/sup\u003e value of 1.0000, and zero error measures (MSE, RMSE, and MAE) show that the values between the expected and actual values are accurate. This good fit is likely to indicate that the intricate non-linear correlations between the tensile strength output and the input parameters have most likely been captured, including factors related to material composition, such as the epoxy content, jute fiber characteristics, and sludge percentage, which have been adequately modelled by the Decision Tree. This is ideal behavior on the training data; however, it can raise concerns regarding possible overfitting, where the model has mastered the training data to an extreme level, considering noise and outliers, and in the process, the model can reduce it capacity to generalize on new data. The model is working on physically viable limits of polymer based composite reinforced with elements of natural fibers and industrial wastes as demonstrated by the tensile strength estimates of between 50 and 55 MPa, and this is equivalent to standard estimates of composite material. The model can generate 100% prediction accuracy and this is strong deterministic relationships among the input parameters and the mechanical properties. This may suggest that elements like compositional ratios, filler particle distribution, and fiber-matrix interface bonding exhibit predictable patterns that the Decision Tree method can accurately map. The tensile behavior of these hybrid composites is probably governed by critical threshold values and interaction effects between factors that the model has identified. The Decision Tree is a potentially useful tool for material design optimization and property prediction within the tested parameter space, as the performance metrics confirm that the chosen features contain enough information to fully describe the tensile strength variations seen in the experimental data.\u003c/p\u003e\n\u003ch3\u003eRandom Forest regression model\u003c/h3\u003e\n\u003cp\u003eThe evaluation metrics presented show (Figure 2(b)) that the Random Forest regression model has a high and reliable prediction performance for tensile strength estimation. The model demonstrates a high level of explanatory power with an R\u003csup\u003e2\u003c/sup\u003e of 0.9374. Together, the error metrics MSE of 1.3403, RMSE of 1.1577, and MAE of 0.6543 verify that the model\u0026apos;s predictions closely match the measured data. The R\u003csup\u003e2\u003c/sup\u003e score of 0.9374 indicates that the input properties of the model account for about 93.74% of the variation in the specimens\u0026apos; tensile strength. This high value indicates that the model has been successful in locating and utilizing the dataset\u0026apos;s underlying patterns and relationships. The magnitude of prediction errors is revealed by the error metrics. The overall prediction and actual tensile strength have an average difference of about 0.65 MPa which is shown by the MAE of 0.6543. No remarkable outliers with severe prediction errors, which is proved by the RMSE of 1.1577 MPa, which is more susceptible to larger errors and as such is moderately low. The visual confirmation of this excellent performance is seen in the almost clustering of the predicted values to the actual trend line in the chart. This is a characteristic of an optimized Random Forest model. A Random Forest is an ensemble algorithm, unlike a single Decision Tree, it creates a number of Decision Trees and aggregates their forecasts. The perfect-score Decision Tree example raised concerns about overfitting, which is obviously addressed by this method. The metrics shown by modeling the complicated, non-linear relationship between input parameters such as material composition, fiber properties, and processing conditions and the tensile strength output without simply remembering the data set indicate that the model has been able to generalize on the training data. In the case of composite materials particularly those made on the basis of polymeric matrices reinforced with natural fibers, the expected tensile strength values that lie between 50 MPa and 58 MPa are physically achievable. The ability of the model to accurately predict within this range proves that the selected features have sufficient details to describe the primary factors influencing the tensile behavior. Because it can accurately forecast mechanical properties based on compositional inputs, the Random Forest model is a very effective and dependable tool for the design and optimization of such composite materials.\u003c/p\u003e\n\u003ch3\u003eGradient Boosting regression model\u003c/h3\u003e\n\u003cp\u003eFor the tensile strength estimation, the Gradient Boosting regression model exhibits remarkable predictive ability, attaining nearly flawless assessment metrics. With comparably low error measures of MSE = 0.0000, RMSE = 0.0001, and MAE = 0.0001, the model achieves an R\u003csup\u003e2\u003c/sup\u003e value of 1.0000. Figure 2(c) shows that the model has learned the fundamental patterns and relationships in the training dataset with exceptionally high precision, leading to low prediction error in all samples. The model accounts for nearly all of the variation in the tensile strength data when the R\u003csup\u003e2\u003c/sup\u003e value is near 1.0. A very close alignment between the projected and actual values is confirmed by the minimal error metrics (MSE, RMSE, and MAE), with the MAE showing an average prediction error of less than 0.0001 MPa. Such a good fit implies that the Gradient Boosting algorithm has been able to model the complex, non-linear relationships and sensitive interaction effects between the tensile strength output and input variables, such as the percentage of sludge filler, epoxy resin content, and jute fiber properties. The high prediction accuracy of the model is among the key features of the Gradient Boosting framework. Gradient Boosting builds trees one after another, each tree is directly trained on the residual (or errors) of the existing ensemble as opposed to bagging algorithms such as Random Forest, which build trees in parallel. This iterative, so-called boosting process may focus on the most difficult cases, resulting in a powerful prediction model that often outperforms traditional ensemble methods. According to the chart, the tensile strength forecasts seem to fall between 48 MPa and 61 MPa, which is in good agreement with usual values for composite materials. For polymer-based composites reinforced with natural fibers and industrial waste components, this shows that the model is functioning within physically reasonable constraints. The performance metrics confirm that the chosen features provide enough information for the model to accurately characterize the differences in tensile strength. The inherent regularization in contemporary Gradient Boosting implementations makes it a highly dependable and useful tool for material design optimization and property prediction within the tested parameter space, even though such high performance needs to be verified on a different test set to definitively rule out overfitting.\u003c/p\u003e\n\u003ch3\u003eXGBoost (Extreme Gradient Boosting) regression model\u003c/h3\u003e\n\u003cp\u003eThe XGBoost (Extreme Gradient Boosting) regression model achieves outstanding assessment metrics and shows state-of-the-art predictive performance for tensile strength prediction (Figure 2(d)). With comparably low error measures of MSE = 0.0000, RMSE = 0.0008, and MAE = 0.0005, the model achieves an R\u0026sup2; value of 1.0000. This shows that the model has nearly zero prediction error across all samples, indicating that it has learned the fundamental patterns and relationships within the training dataset with exceptional precision. Nearly all of the variance in the tensile strength data can be explained by the model when the R\u003csup\u003e2\u003c/sup\u003e value is close to 1.0000. A nearly perfect alignment between the projected and actual values is confirmed by the incredibly low error metrics (MSE, RMSE, and MAE), with the MAE showing an average prediction error of just 0.0005 MPa. This better fit indicates that the XGBoost algorithm has successfully mapped the interaction between input parameters such as epoxy resin content, jute fiber properties, and sludge filler percentage and the tensile strength output in addition to capturing the intricate, non-linear relationships. One distinguishing feature of the XGBoost architecture is the model\u0026apos;s capacity to reach such high prediction accuracy. According to Figure 2(d), the tensile strength predictions seem to fall between 48 and 61 MPa, which is in good agreement with accepted values for composite materials. This suggests that for polymer-based composites reinforced with natural fibers and components from industrial waste, the model\u0026apos;s predictions are limited within physically reasonable limits. The performance measurements confirm that the chosen features offer a comprehensive and adequate explanation for the observed variances in tensile strength. XGBoost\u0026apos;s inherent regularisation and cross-validation capabilities make it an especially reliable and useful tool for the inverse design and optimization of new composite materials, enabling highly accurate prediction of mechanical properties from compositional inputs, even though validation on a hold-out test set is still a crucial step.\u003c/p\u003e\n\u003ch2\u003eComparative performance analysis\u003c/h2\u003e\n\u003cp\u003eThe MSE, RMSE, and MAE values of the four machine learning models Decision Tree, Random Forest, Gradient Boost, and XGBoost that are used to predict the tensile strength of composite materials show (Figure 2(e)) a distinct hierarchy in predictive capability. Although the Decision Tree model provides a simplistic predictive model, the error measures are the highest, implying that this model has a low capacity to extrapolate the complex non-linear relationships between material composition factors such as epoxy content, jute fiber properties, and sludge percentage, and the tensile strength output. This is likely due to the fact that the Decision Tree model is more inclined to overfitting, and the Random Forest model, in contrast, exhibits a significant improvement in performance due to its bagging effect, where multiple trees are constructed simultaneously to reduce variance and decrease overfitting, to achieve much lower error measures and more affordable predictor model. Even more gains in performance are obtained by the Gradient Boost algorithm, which is a sequence of boosting methods in which a new tree corrects the errors of its predecessors. This iterative refinement enables it to capture complex patterns and subtle interaction effects more effectively, resulting in superior accuracy and the lowest error metrics among the conventional models. XGBoost is the most accurate and dependable tool for the inverse design and optimization of these composite materials because of its gradual advancement from a single tree to complex ensemble methods, which confirms that the input features contain enough information to model the tensile behavior.\u003c/p\u003e\n\u003ch2\u003ePredictive performance for flexural strength\u003c/h2\u003e\n\u003ch3\u003eDecision Tree regression model\u003c/h3\u003e\n\u003cp\u003eThe Decision Tree regression model achieves flawless evaluation metrics with an R\u003csup\u003e2\u003c/sup\u003e of 1.0000, MSE of 0.0000, RMSE of 0.0000, and MAE of 0.0000, demonstrating outstanding prediction accuracy for flexural strength estimation.\u003c/p\u003e\n\u003cp\u003eThis Figure 3(a) shows that all specimens, from S0 to S5, have zero prediction error since the model has flawlessly learned and duplicated the underlying patterns in the training dataset. The model explains 100% of the variance in the flexural strength values, as indicated by the R\u003csup\u003e2\u003c/sup\u003e value of 1.0000, and the zero error metrics verify a precise, point-for-point alignment between the anticipated and actual values. This perfect fit indicates that the Decision Tree algorithm has successfully captured the deterministic correlations between the flexural strength output and the input parameters, which are probably processing factors and material composition. Such perfect performance on the training data, however, strongly suggests overfitting, in which the model has probably learned the training dataset too precisely, including its noise and experimental changes, rather than the generalizable underlying principles. As a result, even though the model functions within physically reasonable constraints for the flexural strength of composites, its usefulness is constrained until it can be verified on an independent, unobserved test set to verify its capacity to generalize beyond the precise data on which it was trained.\u003c/p\u003e\n\u003ch3\u003eRandom Forest regression model\u003c/h3\u003e\n\u003cp\u003eStrong assessment metrics, such as a R\u0026sup2; of 0.9292, an MSE of 0.7060, an RMSE of 0.8403, and an MAE of 0.4687, show that the Random Forest regression model has strong and extremely effective predictive performance for evaluating the flexural strength of composite specimens as shown in Figure 3(b). The model successfully explains over 93% of the variance in the flexural strength data, as indicated by the high R\u003csup\u003e2\u003c/sup\u003e value. This shows that the input parameters contain highly useful features for predicting this mechanical attribute. The model\u0026apos;s accuracy is further supported by the low error metrics; the RMSE of 0.8403 indicates that large errors are rare, while the MAE of 0.4687 MPa indicates that, on average, the model\u0026apos;s predictions differ from the actual measured values by less than half a megapascal. This result is typical of the Random Forest\u0026apos;s ensemble structure, which improves generalization on unseen data by combining predictions from several decorrelated decision trees to reduce the danger of overfitting, a common problem with single Decision Trees. The model has successfully learned the underlying non-linear relationships and interaction effects within the material system, as demonstrated by the close alignment between the actual and predicted values across specimens S1 through S5, which is clearly visible in the chart. This makes the model a dependable and effective tool for the design and optimization of composite materials.\u003c/p\u003e\n\u003ch3\u003eGradient Boosting regression model\u003c/h3\u003e\n\u003cp\u003eWith an ideal R\u0026sup2; value of 1.0000 and nearly zero error metrics of MSE = 0.0000, RMSE = 0.0001, and MAE = 0.0001, the Gradient Boosting regression model exhibits theoretically flawless predictive accuracy for evaluating the flexural strength of composite specimens. This is shown in (Figure 3(c)) where the model has almost perfectly captured the underlying patterns and connections in the training dataset, producing predictions that are nearly identical to the actual observed values for every specimen from S0 to S5. The model describes 100% of the variation in the flexural strength data based on the value of an R\u003csup\u003e2\u003c/sup\u003e value of 1.0000 and the marginally small values used to measure the error metrics to almost a perfect congruence of the expected and actual values. Gradient Boosting is characterized by a remarkable fit, and this is due to the fact that it is based on the creation of decision trees in succession. Every new model is uniquely trained to fit the residual errors of the model before it and allows a more complex, non-linear relationship to be represented with an ensemble as well as subtle interaction effects among input parameters being captured with extreme accuracy. The model may have memorized the training data, its noise, and its specific variations in experiments to the extent that its performance on new, unknown data may be compromised. Nevertheless, such a perfect match to the training data is a cause of great concern with regard to overfitting. Hence, prior to the consistent application of this model to the inverse design of new composite materials, its practical applicability and generalizability need to be stringently assessed against a separate test set, even though it is a powerful means of analysis within the parameters space being tested.\u003c/p\u003e\n\u003ch3\u003eXGBoost regression model\u0026nbsp;\u003c/h3\u003e\n\u003cp\u003eHaving an ideal value of R\u003csup\u003e2\u0026nbsp;\u003c/sup\u003e(1.0000) and a low error value of MSE = 0.0000, RMSE = 0.0008, and MAE = 0.0005, the XGBoost regression model presents impressive results in its almost perfect predictive accuracy in the assessment of the flexural strength of composite specimens. Figure 3(d) demonstrates that the model has significantly succeeded in capturing the underlying trends in the training data, and the predictions it makes are almost precise values of its measurements of all the specimens. The very small error values, particularly, the MAE of 0.0005 MPa, demonstrate that the mean error of prediction is not significant, which means that the model is highly accurate. The R\u003csup\u003e2\u003c/sup\u003e value of 1.0000 confirms that the model covers 100% of the variance in data of the flexural strength. This demonstration shows that the XGBoost algorithm has sophisticated functionality in building the relationships and slight interaction effects that are complex and non-linear between the input parameters, and is capable of doing so with higher effectiveness due to integrated L\u003csub\u003e1\u003c/sub\u003e and L\u003csub\u003e2\u0026nbsp;\u003c/sub\u003eregularization to prevent overfitting, a more efficient process of tree-building, and a more advanced treatment of missing data. The regularization mechanisms inherent to the model ensure that it is a highly trustworthy and valuable model in the prediction and optimization of mechanical properties in composite material design although the near-perfect training result needs to be validated on an independent test set to ensure conclusively that the model is generalized.\u003c/p\u003e\n\u003ch3\u003eComparative performance analysis\u003c/h3\u003e\n\u003cp\u003eThe error measures of the four machine learning models provided in Figure 3(e) clearly demonstrate the existence of a strong hierarchy in predicting mechanical properties. The Decision Tree model was found to have the most errors; approximately 0.7 in MSE, 0.84 in RMSE, and 0.47 in MAE, which implies that it has a low predictive ability in generalization and shows a strong overfitting due to simple, single-tree design. The Random Forest model is much more effective, and the error indicators are reduced nearly by half (MSE is approximately 0.35, RMSE is approximately 0.59, and MAE is approximately 0.25). This demonstrates that its ensemble bagging strategy is a good method for reducing variance and enhancing predictive strength. The Gradient Boost model further enhances this accuracy by employing a sequential boosting mechanism, which progressively corrects residual errors to capture more complex patterns in the data resulting in even lower errors of approximately 0.15 for MSE, 0.39 for RMSE and 0.18 for MAE. Ultimately, the XGBoost model has the smallest error measures with MSE approximately 0.05, RMSE approximately 0.22, and MAE approximately 0.08. This is a confirmation of its advanced algorithmic enhancements, including incorporated L1 and L2 regularization and improved learning procedure, which together provide the model with the best and most reliable predictions. XGBoost is the most accurate model for mechanical property prediction in this composite material system, and this gradual decrease in error which will reach its peak in performance by this model represents a significant advantage of advanced ensemble methods.\u003c/p\u003e\n\u003ch2\u003ePredictive performance for impact strength\u003c/h2\u003e\n\u003ch3\u003eDecision Tree regression model\u003c/h3\u003e\n\u003cp\u003eConsidering an ideal R\u0026sup2; value of 1.0000 and zero error metrics (MSE = 0.0000, RMSE = 0.0000, MAE = 0.0000), the Decision Tree regression model shows theoretically perfect predictive accuracy for forecasting the mechanical characteristics of the composite specimens.\u003c/p\u003e\n\u003cp\u003eThis Figure 4(a) shows that the model has learned the training dataset with perfect accuracy, generating predictions for each of the six specimens (S0 through S5) that precisely match the real values. Based on the input parameters, which are the particular material compositions of the specimens, the model explains 100% of the variance in the output characteristic, as indicated by the R\u003csup\u003e2\u003c/sup\u003e value of 1.0000. This perfect fit overwhelmingly suggests serious overfitting even if it implies the Decision Tree has properly mapped the deterministic relationship between the compositional inputs and the mechanical output. Rather of learning a generalizable rule, the model has probably committed all of the training data\u0026mdash;including any experimental noise\u0026mdash;to memory.As a result, even while the model performs flawlessly on this particular set of data, its practical applicability is severely constrained because it would most likely be unable to accurately forecast new, unexplored specimen compositions, which would compromise its dependability for material design and optimization.\u003c/p\u003e\n\u003ch3\u003eRandom Forest regression model\u003c/h3\u003e\n\u003cp\u003eStrong performance measures, such as an R\u003csup\u003e2\u003c/sup\u003e of 0.9517, an MSE of 0.1281, an RMSE of 0.3579, and an MAE of 0.2638 kJ/m\u0026sup2;, show that the Random Forest regression model has exceptional predictive power for forecasting the impact strength of composite specimens. The high R\u003csup\u003e2\u003c/sup\u003e value shows (Figure 4(b)) that the model accounts for about 95.17% of the variance in the impact strength data, indicating that the fundamental factors controlling this mechanical property are successfully captured by the input features, which are probably related to material composition and specimen structure. Random Forest\u0026apos;s ensemble approach, which builds several decision trees via bootstrap aggregation (bagging) and averages their predictions, effectively reduces the overfitting that is frequently connected to individual trees, improving the robustness and applicability of the model. Given the common range of impact strength for composite materials, the model\u0026apos;s predictions differ from the actual impact strength values by just about 0.26 kJ/m\u0026sup2; on average, according to the low error metrics, especially the MAE of 0.2638 kJ/m\u0026sup2;. This finding indicates the ability of the model to capture non-linear and complex relationships and interaction effects in the data which provides it one of the most reliable models in optimizing material formulation and predicting the mechanical behavior of composite systems.\u003c/p\u003e\n\u003cp\u003e.Gradient Boosting\u0026nbsp;regression model\u003c/p\u003e\n\u003cp\u003eThe Gradient Boosting regression model demonstrates an ideal fit to predict the impact strength of the composite specimens with a R\u003csup\u003e2\u003c/sup\u003e = 1.0000 and with zero error values (MSE = 0.0000, RMSE = 0.0000, MAE = 0.0000 kJ/m\u003csup\u003e2\u003c/sup\u003e). The Figure 4(c) shows that the model\u0026apos;s predictions for specimens S0 through S5 had zero average prediction error, no unexplained variance, and perfect, point-for-point alignment with the actual measured impact strength values. The Gradient Boosting algorithm\u0026apos;s sequential learning process, in which each new decision tree in the ensemble is specially built to correct the residual errors of the combined previous trees, is characterized by this faultless performance, which enables it to gradually model the intricate, non-linear relationship between the input parameters and impact strength with ultimate precision on the training data. Instead of learning a generalizable function that would accurately predict the impact strength of novel, untested material compositions, such a perfect result is a classic sign of extreme overfitting, implying that the model has effectively memorized the training dataset including any experimental noise or particular peculiarities of the six specimens. This makes the model an ideal analytical tool for the available data, but without validation on a different test set to verify its capacity to generalize beyond the specific specimens it was trained on, its practical utility for predictive material design is seriously problematic.\u003c/p\u003e\n\u003ch3\u003eXGBoost regression model\u003c/h3\u003e\n\u003cp\u003eUsing an optimal R\u003csup\u003e2\u003c/sup\u003e value of 1.0000, the XGBoost model exhibits nearly perfect prediction accuracy for estimating impact strength, confirming that the model explains 100% of the variance in the target property based on the input data as shown in Figure 4(d). With an RMSE of 0.0008 and MSE and MAE of a minimal 0.0000 and 0.0006 kJ/m\u0026sup2;, respectively, the error metrics show that the predictions are almost exactly the same as the actual measured values across all specimens from S0 to S5, with an average error of only 0.0006 kJ/m\u0026sup2;. XGBoost\u0026apos;s sequential, additive tree-building method, which systematically eliminates residuals from earlier models while controlling for overfitting through built-in L1 and L2 regularization, and its sophisticated regularization algorithms are responsible for its remarkable performance. Their final result is an exceptionally complex model which does not overfit as is common with ideal training scores and represents complex, non-linear relationships and nuanced interaction effects in the data of material composition. As a result, XGBoost is a very dependable tool for the precise prediction and optimization of impact strength in novel composite material formulations since it not only reaches theoretical perfection on the training set but also shows a great potential for robust generalization.\u003c/p\u003e\n\u003ch3\u003eComparative performance analysis\u003c/h3\u003e\n\u003cp\u003eA clear performance hierarchy for forecasting the mechanical properties of composite specimens is revealed by comparing the error metrics of the four machine learning models as shown in Figure 4(e). \u0026nbsp;Due to a high tendency for overfitting from its single-tree structure, the Decision Tree model has the highest error values. Its metrics, which are probably near the top of the scale at 0.30 for MSE, 0.55 for RMSE, and 0.25 for MAE, highlight its fundamental limitation in generalizing beyond the training data. The Random Forest model, which uses ensemble bagging, reduces these errors by roughly half, to about 0.15 in MSE, 0.39 in RMSE, and 0.12 in MAE. This is the confirmation that the model is stronger and better in ability to define the underlying data patterns using the aspect of variance reduction. The Gradient Boost model builds on this by applying a sequential boosting algorithm to the error again to get it down to approximately 0.07 on MSE, 0.26 on RMSE and 0.08 on MAE. This demonstrates that it is a highly advanced method of modeling complex non-linear interactions through the correction of prediction residuals during a sequence of steps. XGBoost model eventually attains the lowest error rates, the scores go down to nearly zero- 0.02 in MSE, 0.14 in RMSE and 0.03 in MAE. \u0026nbsp;It is effective because of its very efficient learning algorithm and advanced regularization which in combination enable the model to possess unmatched predictive precision and opens up to its status as the optimal model to possess high-reliability predictability of mechanical properties in composite material systems.\u003c/p\u003e\n\u003ch2\u003eOverall feature importance Heatmap\u003c/h2\u003e\n\u003cp\u003eThe feature importance analysis provides clear-cut mechanistic roles of each composite constituent in the control of mechanical performance (Figure 5(a)). The content of epoxy shows the highest effect on all properties, especially in terms of impact strength, and the value of importance of 0.746. This large dominance demonstrates the important role of the matrix in the energy absorption and crack resistance, in which continuous and tough epoxy phase improves the ductile and the stress distribution of the interfaces, making the impact resistance increase dramatically between 13.352 kJ/m\u003csup\u003e2\u003c/sup\u003e in pure epoxy to 18.026 kJ/m\u003csup\u003e2\u003c/sup\u003e in the composite of the optimum. In tensile and flexural strength, epoxy is the most affecting variable (0.480 and 0.471, respectively), which is expected given that this is the primary load-bearing medium, which upholds the tensile strength to rise to 61.84 MPa and flexural strength to 31.81 MPa in S4 composite.\u003c/p\u003e\n\u003cp\u003eJute fiber demonstrates high significance on tensile (0.341) and flexural strength (0.370), which can be attributed to the fact that the fiber reinforces the matrix through fiber bridging and load transfer, and the fiber indicates a low contribution towards impact strength (0.041) meaning that during high-rate loading, the bonding between the fiber and the matrix can be damaged prematurely, which restricts the energy absorption. Although Sludge content has the smallest effect on strength properties (0.179 tensile and 0.159 flexural), it has a significant effect on impact strength (0.213), meaning that the hard sludge particles contribute to crack deflection and localized plastic deformation, though overloading the material above 20 wt.% reduces interfacial adhesion and boosts brittleness as seen by the deterioration of mechanical properties at 25 wt. % sludge.\u003c/p\u003e\n\u003ch2\u003eCorrelation matrix\u003c/h2\u003e\n\u003cp\u003eThe correlation matrix shown in Figure 5(b) indicates that negativities of composition are strong as the correlations between epoxy, jute, and sludge are high with negative correlation of -0.92, -0.90; this can be attributed to the fact that an increase in the percent composition causes a reduction in the others as dictated by the formulation constraint: the total weight percentage has to be 100. This inherent trade-off is a direct effect on behavior of mechanical properties. All of the three mechanical properties, tensile strength (-0.98), flexural strength (-0.96), and impact strength (-0.96) have very strong negative relationships with sludge content, which confirms that above the optimal sludge loading (20 wt.%) the mechanical properties are reduced by agglomeration of particles, lack of adhesion, and the increase in the content of voids. On the other hand, tensile (0.98), flexural (0.96), and impact strength (0.96) exhibit very strong positive correlations with epoxy content, which highlights the importance of polymer matrix in the maintenance of structural integrity, transfer of stress, and toughness provided by good interfaces with both jute fibers and sludge particles. Another good correlation can be found between jute fiber content and tensile (0.92), flexural strength (0.95) which indicates that jute fiber can reinforce the material through the mechanism of fiber-matrix interaction and load bearing, but the correlation between jute fiber content and impact strength is moderate (0.88), indicating that the role of jute fiber is not as dominant as that of the epoxy matrix under dynamic loading. These correlation trends indicate that there is an undeniable mechanistic trade-off that the more epoxy used, the better the mechanical strength achieved, but the more sludge used, though it is good in hardness and density, the more it negatively affects the strength properties beyond a certain optimal level, and this is the way formulation strategies can be pursued towards balanced performance in sustainable hybrid composites.\u003c/p\u003e\n\u003ch2\u003eOptimization of composite composition\u003c/h2\u003e\n\u003cp\u003eThe composite formula (S4) was optimized to reach high mechanical characteristics in comparison to the neat epoxy base, and the improved mechanical properties of the composite formula can be attributed to the synergistic reinforcement processes of the jute fiber and LD sludge (Figure 5(c)). The flexural strength displayed maximum improvement of 41.8 which showed an increase of 22.44 MPa to 31.81 MPa under bending due to high transfer of the load and bridging effect of the jute fibers which in turn reduced the concentration of stress and also delayed the formation of a crack, which is attributed to enhanced interfacial adhesion due to good dispersion of the sludge particles. It has increased tensile strength of 28.8% with the sludge increasing tensile strength of 48 MPa to 61.84 MPa; this is mainly because of the reinforcing nature of jute fibers, which bear the axial loads and the ability of sludge to fill up the micro voids, which in combination with both increase the continuity and resist tensile deformation of the matrix. Impact strength was enhanced by 35.0% to 18.026 kJ/m\u003csup\u003e2\u003c/sup\u003e because of energy dissipation mechanisms (fiber pull-out, crack deflection around sludge particles, and matrix micro-cracking) leading to a significant increase in toughness. These percentage improvements represent the positive incorporation of natural fiber and industrial waste filler, with the jute being the main form of reinforcement, the sludge modifying microstructure and interfacial characteristics, and the epoxy matrix being the type of reinforcement that bonds the hybrid composite system, with the resulting tensile, flexural, and impact properties to be balanced.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThe current research demonstrated successful fabrication of high performance and eco-friendly epoxy composites using jute fiber and LD sludge as hybrid reinforcements. The experimental findings indicate that the S4 formulation (20 wt.% jute, 20 wt.% sludge) provides the best mechanical performance, with tensile strength increased from 48 MPa to 61.84 MPa (28.8%), flexural strength increased from 22.44 MPa to 31.81 MPa (41.8%), and impact strength increased from 13.352 kJ/m\u003csup\u003e2\u003c/sup\u003e to 18.026 kJ/m\u003csup\u003e2\u003c/sup\u003e (35%). It is suggested that these gains are due to synergistic reinforcement mechanisms such as good stress transfer, better fiber-matrix adhesion and deflection of cracks by sludge particles. At the maximum filler loading (25 wt.%), mechanical properties are reduced, and it is essential to have a balanced incorporation of fillers. The application of machine learning models, specifically XGBoost, yielded a highly precise predictive model (R\u0026sup2; = 1.000, MAE = -0.0005 MPa) for predicting mechanical behavior using compositional data, thereby facilitating effective material optimization. The study confirms a dual waste-valorisation and data-driven design approach, which is part of the formation of sustainable composites for structural and industrial use.\u003c/p\u003e \u003cp\u003eThis study demonstrates that industrial waste-filled EP-JF composites can be utilized as low-cost materials, contributing to sustainable material development. The optimised composite system exhibits enhanced mechanical properties, making it a suitable candidate for load-bearing and semi-structural applications such as interior automotive panels, lightweight housings, brackets, enclosures, partition boards, furniture components, building panels, and other structural components where moderate strength, stiffness, and durability are required.\u003c/p\u003e \u003cdiv id=\"Sec29\" class=\"Section2\"\u003e \u003ch2\u003eLimitations of this work\u003c/h2\u003e \u003cp\u003eThis study is limited by the narrow compositional range investigated and the relatively small dataset used for the machine learning model, which may restrict broader generalization. The work focused mainly on mechanical properties, without a detailed evaluation of long-term durability, thermal behavior, or large-scale manufacturing feasibility. Further studies with expanded compositions, larger datasets, and extended performance testing are needed to strengthen practical applicability\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e\n\u003cp\u003e\u003cem\u003eNot Applicable.\u003c/em\u003e\u003c/p\u003e\n\u003ch2\u003ePatient consent statement\u003c/h2\u003e\n\u003cp\u003e\u003cem\u003eNot Applicable.\u003c/em\u003e\u003c/p\u003e\n\u003ch2\u003eConsent to publish\u003c/h2\u003e\n\u003cp\u003e\u003cem\u003eNot applicable.\u003c/em\u003e\u003c/p\u003e\n\u003ch2\u003eFunding statement\u003c/h2\u003e\n\u003cp\u003eThis research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Whereas Open access funding will be provided by Manipal Academy of Higher Education, Manipal\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eAbhilash Purohit contributed to the methodology, wrote the original draft, conducted experiments, and performed data analysis. S. Sathees Kuma contributed to writing the original draft, experiments, and data analysis. Pravat Ranjan Pati contributed to conceptualization, writing, and data curation. Arvind Kumar contributed to conceptualization and provided funding for the experiment. Suresh Palanimuthu contributed to supervising the experiment and provided funding for the experiment. Byomakesh Mahapatra contributed to validation, reviewing, and editing\u003c/p\u003e\n\u003ch2\u003eAcknowledgement\u003c/h2\u003e\n\u003cp\u003eThe article is supported by MAHE OA funding\u003c/p\u003e\n\u003ch2\u003eData Availability\u003c/h2\u003e\n\u003cp\u003eThe data used and/or analyzed during the current study are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eMohanty AK, Misra M, Drzal LT. Sustainable bio-composites from renewable resources: opportunities and challenges in the green materials world. \u003cem\u003eJ Polym Environ\u003c/em\u003e 2002; 10: 19\u0026ndash;26.\u003c/li\u003e\n\u003cli\u003eOchi D, Barbieri D, Reis AF, et al. Chapter 1 - Agro-industrial waste as fillers for green composites. In: Inamuddin, Altalhi T, Alrooqi A (eds) \u003cem\u003eGreen Sustainable Process for Chemical and Environmental Engineering and Science\u003c/em\u003e. Elsevier, pp. 1\u0026ndash;26.\u003c/li\u003e\n\u003cli\u003ePugar D, Ćurković L, Gabelica I, et al. Potential Use of Jarosite Industrial Waste in Developing Hybrid Composites. \u003cem\u003eSustain\u003c/em\u003e; 16.\u003c/li\u003e\n\u003cli\u003eRamesh M, Palanikumar K, Reddy KH. Plant fibre based bio-composites: Sustainable and renewable green materials. \u003cem\u003eRenew Sustain Energy Rev\u003c/em\u003e 2017; 79: 558\u0026ndash;584.\u003c/li\u003e\n\u003cli\u003eSabeel Ahmed K, Khalid SS, Mallinatha V, et al. Dry sliding wear behavior of SiC/Al2O3 filled jute/epoxy composites. \u003cem\u003eMater Des\u003c/em\u003e 2012; 36: 306\u0026ndash;315.\u003c/li\u003e\n\u003cli\u003eGoriparthi BK, Suman KNS, Mohan Rao N. Effect of fiber surface treatments on mechanical and abrasive wear performance of polylactide/jute composites. \u003cem\u003eCompos Part A Appl Sci Manuf\u003c/em\u003e 2012; 43: 1800\u0026ndash;1808.\u003c/li\u003e\n\u003cli\u003eDas B, Prakash S, Reddy PSR, et al. An overview of utilization of slag and sludge from steel industries. \u003cem\u003eResour Conserv Recycl\u003c/em\u003e 2007; 50: 40\u0026ndash;57.\u003c/li\u003e\n\u003cli\u003eReghunadhan A, Datta J, Jaroszewski M. Polyurethane glycolysate from industrial waste recycling to develop low dielectric constant , thermally stable materials suitable for the electronics. \u003cem\u003eArab J Chem\u003c/em\u003e 2020; 13: 2110\u0026ndash;2120.\u003c/li\u003e\n\u003cli\u003eVigneshwaran S, Uthayakumar M, Arumugaprabu V. Potential use of industrial waste-red mud in developing hybrid composites: A waste management approach. \u003cem\u003eJ Clean Prod\u003c/em\u003e 2020; 276: 124278.\u003c/li\u003e\n\u003cli\u003eVinod A, Tengsuthiwat J, Gowda Y, et al. Jute/Hemp bio-epoxy hybrid bio-composites: Influence of stacking sequence on adhesion of fiber-matrix. \u003cem\u003eInt J Adhes Adhes\u003c/em\u003e 2022; 113: 103050.\u003c/li\u003e\n\u003cli\u003eTian J, Li C, Xian G. Reciprocating friction and wear performances of nanometer sized-TiO2 filled epoxy composites. \u003cem\u003ePolym Compos\u003c/em\u003e 2021; 42: 2061\u0026ndash;2072.\u003c/li\u003e\n\u003cli\u003eHsissou R, Seghiri R, Benzekri Z, et al. Polymer composite materials: A comprehensive review. \u003cem\u003eCompos Struct\u003c/em\u003e 2021; 262: 113640.\u003c/li\u003e\n\u003cli\u003eJin F-L, Li X, Park S-J. Synthesis and application of epoxy resins: A review. \u003cem\u003eJ Ind Eng Chem\u003c/em\u003e 2015; 29: 1\u0026ndash;11.\u003c/li\u003e\n\u003cli\u003eSarikaya E, \u0026Ccedil;allioğlu H, Demirel H. Production of epoxy composites reinforced by different natural fibers and their mechanical properties. \u003cem\u003eCompos Part B Eng\u003c/em\u003e 2019; 167: 461\u0026ndash;466.\u003c/li\u003e\n\u003cli\u003eSanjay MR, Madhu P, Jawaid M, et al. Characterization and properties of natural fiber polymer composites: A comprehensive review. \u003cem\u003eJ Clean Prod\u003c/em\u003e 2018; 172: 566\u0026ndash;581.\u003c/li\u003e\n\u003cli\u003eKumar S, Saha A. Graphene nanoplatelets/organic wood dust hybrid composites: physical, mechanical and thermal characterization. \u003cem\u003eIran Polym J\u003c/em\u003e 2021; 30: 935\u0026ndash;951.\u003c/li\u003e\n\u003cli\u003eVengadesan E, Arunkumar T, Muralidharan S, et al. Hybrid bio-composites reinforced with natural wood saw dust and eco-friendly graphite: evaluation of physical, mechanical, and thermal properties. \u003cem\u003eFibers Polym\u003c/em\u003e 2025; 26: 833\u0026ndash;854.\u003c/li\u003e\n\u003cli\u003eNurazzi NM, Harussani MM, Aisyah HA, et al. Treatments of natural fiber as reinforcement in polymer composites-a short review. \u003cem\u003eFunct Compos Struct\u003c/em\u003e 2021; 3: 24002.\u003c/li\u003e\n\u003cli\u003eGupta MK, Srivastava RK. Mechanical Properties of Hybrid Fibers-Reinforced Polymer Composite: A Review. \u003cem\u003ePolym - Plast Technol Eng\u003c/em\u003e 2016; 55: 626\u0026ndash;642.\u003c/li\u003e\n\u003cli\u003eThakur VK, Kessler MR. Self-healing polymer nanocomposite materials: A review. \u003cem\u003ePolymer (Guildf)\u003c/em\u003e 2015; 69: 369\u0026ndash;383.\u003c/li\u003e\n\u003cli\u003ePurohit A, Satapathy A. Development and characterization of epoxy-based composites filled with Linz--Donawitz sludge. \u003cem\u003eJ Compos Mater\u003c/em\u003e 2017; 51: 899\u0026ndash;911.\u003c/li\u003e\n\u003cli\u003ePurohit A, Satapathy A. Processing, characterization, and parametric analysis of erosion behavior of epoxy-LD sludge composites using T aguchi technique and response surface method. \u003cem\u003ePolym Compos\u003c/em\u003e 2018; 39: E2283--E2297.\u003c/li\u003e\n\u003cli\u003ePurohit A, Satapathy A. Mechanical and wear characteristics of epoxy composites filled with industrial wastes: a comparative study. In: \u003cem\u003eIOP Conference Series: Materials Science and Engineering\u003c/em\u003e. 2017, p. 12019.\u003c/li\u003e\n\u003cli\u003ePurohit A, Swain PTR, Patnaik PK. Mechanical and sliding wear characterization of LD sludge filled hybrid composites. \u003cem\u003eMater Today Proc\u003c/em\u003e 2020; 26: 1654\u0026ndash;1659.\u003c/li\u003e\n\u003cli\u003ePurohit A, Tripathy V, Mishra SK, et al. Mechanical and Tribo-performance Analysis of Linz Donawitz Sludge-Filled Glass\u0026ndash;Epoxy Composites using Taguchi Experimental Design. \u003cem\u003eJ Inst Eng Ser E\u003c/em\u003e 2022; 35: 1\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003eKumar SS, Shyamala P, Pati PR. Machine learning algorithms to predict the tensile strength of novel composite materials. \u003cem\u003eNext Mater\u003c/em\u003e 2025; 9: 101105.\u003c/li\u003e\n\u003cli\u003eBeldar P, Kadbhane S. Utilizing machine learning to forecast mechanical characteristics of NaOH-Treated jute fiber reinforced composite materials. \u003cem\u003eMater Lett\u003c/em\u003e 2024; 377: 137411.\u003c/li\u003e\n\u003cli\u003eMilad A, Hussein SH, Khekan AR, et al. Development of ensemble machine learning approaches for designing fiber-reinforced polymer composite strain prediction model. \u003cem\u003eEng Comput\u003c/em\u003e 2022; 38: 3625\u0026ndash;3637.\u003c/li\u003e\n\u003cli\u003eYoussefi M, Safaie B. The Study on the Mechanical Properties of Multi-walled Carbon Nanotube/Polypropylene Fibers. \u003cem\u003eJ Inst Eng Ser E\u003c/em\u003e 2018; 99: 37\u0026ndash;42.\u003c/li\u003e\n\u003cli\u003eRahman ANMM, Ahmed F, Alimuzzaman S, et al. Enhancement of Thermo-mechanical Properties of Okra Fiber by Photografting Technique. \u003cem\u003eJ Inst Eng Ser E\u003c/em\u003e 2021; 102: 45\u0026ndash;59.\u003c/li\u003e\n\u003cli\u003eMahapatra SS, Patnaik A. Study on mechanical and erosion wear behavior of hybrid composites using Taguchi experimental design. \u003cem\u003eMater \\\u0026amp; Des\u003c/em\u003e 2009; 30: 2791\u0026ndash;2801.\u003c/li\u003e\n\u003cli\u003eErdo\\ugdu YE, Korkmaz EE, Temiz \\cSemsettin. Effect of graphene nanoplatelet filling on mechanical properties of natural fiber reinforced polymer composites. \u003cem\u003eMater Test\u003c/em\u003e 2021; 63: 322\u0026ndash;328.\u003c/li\u003e\n\u003cli\u003eMirzamohammadi S, Eslami-Farsani R, Ebrahimnezhad-Khaljiri H. The experimental assessment of carbon nanotubes incorporation on the tensile and impact properties of fiber metal laminate fabricated by jute/basalt fabrics-aluminum layers: as hybrid structures. \u003cem\u003eProc Inst Mech Eng Part C J Mech Eng Sci\u003c/em\u003e 2023; 237: 1877\u0026ndash;1886.\u003c/li\u003e\n\u003cli\u003eOlhan S, Antil B, Behera BK. Low-velocity impact and quasi-static post-impact compression analysis of woven structural composites for automotive: Influence of fibre types and architectural structures. \u003cem\u003eCompos Struct\u003c/em\u003e 2025; 352: 118676.\u003c/li\u003e\n\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":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Composites, Industrial Waste, Jute Fiber, Machine Learning, Mechanical Characterization","lastPublishedDoi":"10.21203/rs.3.rs-8903744/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8903744/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study investigates the development, characterization, and predictive modelling of sustainable epoxy composites reinforced with jute fiber and Linz-Donawitz (LD) sludge of industrial waste. Specimens were made using different LD sludge content (0\u0026ndash;25 wt.%) and the constant jute fiber loading of 20 wt.%. The mechanical testing showed that the best composition was 60 wt.% epoxy, 20 wt.% jute, 20 wt.% LD sludge that demonstrated considerable gains in tensile strength (up to 61.84 MPa, an improvement of 28.8%), flexural strength (up to 31.81 MPa, an improvement of 41.8%), and impact strength (up to 18.026 kJ/m\u003csup\u003e2\u003c/sup\u003e). After 20 wt.% sludge, the mechanical performance decreased because of interfacial defects. Machine learning models, namely Decision Tree, Random Forest, Gradient Boosting, and XGBoost were implemented in Google Colab to predict mechanical properties from compositional inputs. XGBoost showed better predictive performance as its error measures were close to zero (MAE\u0026thinsp;=\u0026thinsp;0.0005 MPa tensile strength), validating its utility for inverse material design and the optimization of sustainable hybrid composites. The findings highlight a combined waste-valorisation and data-driven design approach that supports the development of sustainable composites for structural and industrial use.\u003c/p\u003e","manuscriptTitle":"Machine Learning Approach for Mechanical Property Assessment of Industrial Waste-Filled Epoxy-Jute Composites","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-26 19:42:19","doi":"10.21203/rs.3.rs-8903744/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-30T09:19:33+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-29T08:39:36+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-29T04:53:26+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-22T23:52:14+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"283901613067238956758850463994331745341","date":"2026-03-20T15:01:11+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"214452513454451030716977405201400881026","date":"2026-03-20T12:11:53+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"239948536570380190358517723753032095076","date":"2026-03-20T11:52:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"114820761208453160105450322720004856810","date":"2026-03-01T22:54:37+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"248687844988928218380256815186670595175","date":"2026-03-01T08:50:15+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-24T21:51:19+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"209634906909214772641092453115041517653","date":"2026-02-24T21:24:11+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"21568890010829403147511377577869823952","date":"2026-02-24T09:24:46+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-24T08:23:44+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-24T08:20:32+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-02-24T03:47:49+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-20T11:55:33+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-02-20T11:50:05+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d37b6cc0-b9e2-40eb-b29a-b0726dee7c66","owner":[],"postedDate":"February 26th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":63476302,"name":"Physical sciences/Engineering"},{"id":63476303,"name":"Physical sciences/Materials science"}],"tags":[],"updatedAt":"2026-05-12T03:55:55+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-26 19:42:19","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8903744","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8903744","identity":"rs-8903744","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}