Hybrid Comparative Modeling of ANN and SVM for Accurate Machining Performance Prediction of AlSiC MMC

Hybrid Comparative Modeling of ANN and SVM for Accurate Machining Performance Prediction of AlSiC MMC | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Hybrid Comparative Modeling of ANN and SVM for Accurate Machining Performance Prediction of AlSiC MMC AMOL ASHOK CHAVAN, Dadarao Raut This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7378480/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract This study presents a hybrid machine learning model that combines Artificial Neural Networks (ANN) and Support Vector Machines (SVM) to model critical machining performance parameters such as surface roughness (Ra), material cutting force (Fc) as well as material removal rate (MRR) in the machining of aluminum silicon carbide (AlSiC) metal matrix composite by turning. A synthetic data set consisting of literature sources validated on a model-to-case basis was employed to train both individual models and the architecture itself and its performance was assessed through the statistical metrics (R2, RMSE, MAE) and the visual inspection of the data (parity plots, residual histograms). ANN was proven to be more accurate in modeling nonlinear surface roughness behavior, and SVM was more accurate in estimating MRR since it had the capacity to have a smooth generalization. The inverse-RMSE weighted hybrid ensemble was always better at the task than the two standalone models and provided lower RMSE values with the minima of 0.055 µm on Ra and 2.361 N on Fc. These findings validate the soundness and applicability of the hybrid model as a digital twin to CAM-based process leads the way with tool-life management and efficiency maximization. The applications in adaptive precision machining such as real-time validation and incorporation of the other sensor variables will be done in the future. AlSiC composite hybrid modeling surface roughness prediction cutting force material removal rate ANN SVM intelligent machining digital twin Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 1 Introduction Metal Matrix Composites (MMCs) are drawing increasing attention in the current landscape of advanced manufacturing due to their exceptional mechanical, thermal, and tribological properties. Among these, Aluminum-Silicon Carbide (AlSiC) composites have emerged as particularly attractive materials because they offer a unique combination of lightweight strength, wear resistance, and thermal stability, making them suitable for demanding applications in the automotive, aerospace, electronics, and defense sectors. Miracle ( 2013 ) highlighted the importance of MMCs like AlSiC in aerospace and space systems, owing to their high stiffness-to-weight ratio and superior thermal performance. The mechanical viability of hybrid aluminum MMCs for automotive components has also been validated in recent work by Nirala et al. (2022), further establishing their utility under dynamic loading conditions. Given the modern industrial emphasis on lightweight and high-performance materials, AlSiC composites have become essential for overcoming operational challenges in extreme environments. However, despite these advantages, machining of AlSiC poses significant difficulties. The hard ceramic reinforcement of SiC particles causes rapid tool wear, inferior surface finish, and non-linear force patterns during machining. While non-conventional techniques like Electrical Discharge Machining (EDM) and Electrochemical Machining (ECM) offer promise Hinduja & Li (2012), they are limited by surface quality, high tooling costs, and material conductivity issues. In conventional turning and milling, achieving precision in AlSiC machining remains a persistent challenge, as the complex interactions between process parameters (cutting speed, feed rate, and depth of cut) and output variables (surface roughness, cutting force, and material removal rate) are not adequately captured by traditional empirical models. To address this, there is a growing need for data-driven, smart modeling approaches for prediction and optimization. Soft computing techniques, including machine learning and hybrid intelligence methods, have shown greater flexibility and accuracy in modeling the nonlinear behaviors of MMC machining processes, as emphasized by Laghari et al. ( 2019 ). This research is motivated by the necessity to develop scalable and generalizable predictive frameworks capable of handling multiple performance metrics simultaneously. Prior studies have mostly focused on single-response modeling or standalone ML algorithms without leveraging ensemble methods. Given the rising importance of digital twins and Industry 4.0 paradigms, there is now a strong incentive to design hybrid models that support real-time, data-driven decision-making in precision machining. The use of ML in manufacturing, particularly for predicting and controlling the machining performance of Al-based composites, has grown significantly. ANNs have effectively modeled complex nonlinear relationships, as demonstrated by Karim et al. ( 2018 ) and Saini et al. ( 2021 ) in modeling Ra and MRR during dry and MQL-assisted turning of SiC-reinforced Al composites. Reddy et al. ( 2025 ) used ANN optimization in a hybrid composite (Al-Cu-SiC-GNP), achieving enhanced machinability and mechanical behavior. SVMs, known for their generalization ability with small datasets, have also been effectively applied by Saikrupa et al. ( 2025 ), Sekhar et al. ( 2021 ), and Ulas et al. ( 2020 ) in various machining contexts, including cryogenic, microstructure, and wire-EDM machining of AlSiC-based composites. Benchmarking studies have increasingly turned to hybrid approaches combining ANNs with optimization algorithms such as Genetic Algorithms (GA) and Swarm Intelligence. For example, Selvarasu and Jayakumar (2021) integrated deep neural networks with Salp Swarm Optimization (SSO) to forecast machinability in granite/SiC-Al composites, while Kalita et al. ( 2023 ) developed a MOALO-MODA ensemble model for multi-objective optimization in MMC turning. Comparative studies by UCAR and KATI (2020) found that hybrid and ensemble models outperform standalone ML approaches in terms of overfitting, extrapolation, and interpretability issues. Research by Weichert et al. ( 2019 ) has further validated the importance of ML for optimizing machining performance in high-dimensional, nonlinear environments. Despite these advances, existing studies often lack multi-output prediction capabilities, fail to integrate the strengths of various ML techniques, or overlook consistency across performance metrics such as Ra, Fc, and MRR. This opens a clear opportunity for developing hybrid ensemble models that can capture the complex interdependencies between machining parameters and outcomes across a broad operational range. Although ANN and SVM have shown promise individually, their combined hybrid use, particularly for simultaneous multi-output prediction in AlSiC machining, has not been sufficiently explored. Most prior research is limited to single-response modeling, cross-validation, or isolated optimization, neglecting the benefits of hybrid ensembles in reducing prediction variance and improving interpretability. To address this gap, the current study introduces a novel comparative and hybrid modeling framework that combines ANN and SVM to predict key machining responses—surface roughness (Ra), cutting force (Fc), and material removal rate (MRR). The specific objectives are to: (1) develop and evaluate individual ANN and SVM models for multi-output machining performance prediction, (2) design a hybrid ensemble model using inverse-RMSE-based dynamic weighting of ANN and SVM predictions, (3) assess and compare model performances using statistical metrics such as R², RMSE, and MAE, and (4) visualize and interpret prediction accuracy through parity plots, residual distributions, and feature-wise data representations. The contribution of this research lies in its development of a lightweight yet effective hybrid ensemble model that enhances prediction accuracy, stability, and interpretability in AlSiC machining scenarios. This model, validated using literature-based datasets, is well-suited for modeling realistic turning processes under a broad spectrum of machining conditions. The findings have significant implications for smart machining analytics, enabling more efficient process planning, tool wear monitoring, and adaptive manufacturing of composite materials. Ultimately, this study contributes to the integration of artificial intelligence in machining processes, supporting the development of more intelligent, responsive, and data-driven production systems in advanced manufacturing environments. 2 Experiemental Details Metal matrix composites (MMCs), in particular, Aluminium-Silicon Carbide (AlSiC) types, are a particular challenge to machine as they have a heterogeneous microstructure and SiC reinforcements are abrasive. Such composites are finding more high-performance applications in the aerospace, automotive and defence industries where high strength to weight ratio, thermal resistance and dimensional stability are paramount. But, the abrasive particles of SiC cause quick wear and surface integrity problems in the conventional machining process. Thus, proper predictive modelling of the machining process parameters including surface roughness (Ra), cutting force (Fc), and material removal rate (MRR) becomes crucial in the optimization of the process as well as quality control. To allow such predictions, a synthetic, realistic dataset that represents a wide variety of machining conditions was created and analyzed as explained in the following subsection: 2.1 Materials and Process Overview The research takes into account the AlSiC-based MMCs as the main material of workpieces. Such composites usually use an aluminum matrix, strengthened with silicon carbide particles (usually 1020 percent, by volume), to increase mechanical and tribological performance. In this study, the attention was given to turning operations because they are widely used in industries and the benchmarks of performance are provided in the literature. The simulation based on accepted machining behavior and parameters range of earlier literature including Radhakrishnan et al. ( 2011 ), Muthukrishnan & Davim ( 2009 ), and Dhavamani & Alwarsamy ( 2012 ) which optimized machining parameters of Al-SiC composites using genetic algorithms, was intended to model three key responses- Ra, Fc and MRR under different cutting regimes. The following are the process parameters to be simulated: Cutting Speed (Vc): 100–250 m/min Feed Rate (f): 0.05–0.3 mm/rev Depth of Cut (d): 0.5–2.0 mm These ranges reflect realistic values adopted in machining Al/SiC composites under dry and minimum quantity lubrication (MQL) conditions. 2.2 Dataset Construction and Parameters To prevent the logistical and financial overheads of large-scale physical experimentation and satisfy scientific rigor, a dataset was created based on a simulation-based design-of-experiments (DOE) approach. The parameter windows of the simulated dataset were chosen in accordance with the validated parameters in several experimental studies Karim et al (2020) and Chawla et al ( 2021 ). Particularly, Dabade & Jadhav ( 2016 ) emphasized the applicability of hot machining strategies in enhancing the surface integrity of AlSiC composites in diverse turning conditions, thus informing the applicable limits of parameters in the current research. Also, the multi-response optimization strategy suggested by Miladinovic et al. (2024) to apply to Al-Si composites gave an understanding of choosing reliable parameter combinations that would be applicable in ANN-based modeling. The design space was sampled uniformly on the inputs (Vc, f, d) in order to have the points evenly distributed in the design space, and not clustered on parameters which is known to bias training of machine learning models. They produced 60 unique data points or different machining conditions with related predicted values of Ra, Fc and MRR. 2.3 Simulated Trial Generation The simulated trials were generated by full-factorial combinations of the input parameters, that is, a complete coverage of both conservative and aggressive machining zones. The approach guarantees the downstream applications of generalizable predictive modeling. A 3D scatter plot (Fig. 1 ) was created to graphically confirm that the input conditions have been evenly distributed. The plot verifies the spatial consistency of the data in all three parameters which are cutting speed (Vc), feed rate (f) and depth of cut (d). This coverage is essential in reducing the extrapolation errors in training and testing of ML models. The proposed simulation methodology is justified by the previous benchmark experiments like Stojanovic et al. (2017), which managed to apply virtual trials to the assessment of the MMC machining behavior. 2.4 Data Characteristics and Correlation Insights The third dataset consisted of three input variables which were independent and three output variables which were dependent. It was necessary to develop a correlation heatmap (Fig. 2 ) to measure the linear relationships between the variables to guarantee data quality and suitability to apply supervised machine learning. Key observations from the heatmap include: Feed rate had a strong positive correlation with MRR (0.59), and Ra (0.86), indicating that an increase in feed results in an increase in removal rate and surface irregularity-as indicated by machining theory. Depth of cut correlated strongly with Fc (0.79), supporting literature findings that higher depths cause increased tool-workpiece interaction and force Mohan et al., ( 2008 ). Cutting speed had a moderate negative correlation with both Ra and Fc, affirming that higher speeds can reduce force and improve finish by minimizing chip load per revolution Chen et al., ( 2024 ). The balanced distribution, the lack of multicollinearity, the domain-consistent correlations, and the lack of correlations between the variables verify the credibility of the dataset to be used in regression-based machine learning to predict the machining responses of AlSiC composites. 3 Methodology The rigorous methodological approach is needed in order to create an interpretable and accurate predictive framework of machining of AlSiC composites. Since the nonlinear relationship between machining parameters and output responses, to include surface roughness (Ra), cutting force (Fc), and material removal rate (MRR), could not be described with simple models, a mixture of advanced machine learning methods and hybrid modelling was embraced. The section gives the step-by-step guidelines to be used- that is, data pre-processing, model evaluation. The complete landscape of the end-to-end structure is visualized at Fig. 3 and Fig. 4 , with each stage of the methodological process described below. 3.1 Data Preprocessing and Feature Scaling The basis of performance of machine learning lies in high-quality input data. The data used consists of three independent machining variables, which include cutting speed (Vc), feed rate (f), and depth of cut (d) and three dependent machining performance variables, which include Ra, Fc and MRR. To guarantee uniform convergence as well as scale invariance among features, it was decided to normalize z-scores of all the input variables: $$\:{x}_{scaled}=\:\frac{x-\:\mu\:}{\sigma\:}$$ where µ is the mean and σ is the standard deviation of the feature. This process adapts the features, so that they are centered and have a unit variance, which allows optimal performance of both the kernel-based methods and the neural networks. After normalization, the data was randomly split into training (85%) and testing (15%) sets by a stratified sampling to maintain the base distributions to train an unbiased model without any bias in evaluation of model. 3.2 Artificial Neural Network (ANN) Model Design An Artificial Neural Network (ANN) comprising of Tensor Flow was developed to capture the nonlinear interactions between input and output variables. The architecture was well balanced to make it possible to perform multi-output regression modelling between the complexity and the generalization of the architecture. Previous research to promote the use of ANN in this undertaking includes Chandrasekaran & Devarasiddappa ( 2014 ) that proved that it is effective in predicting surface roughness when grinding Al-SiCp metal matrix composites. Also, an ANN and Response Surface Methodology (RSM), was comparatively analyzed by Rajbongshi & Sarma ( 2019 ), with the conclusion that ANN performed better by the performance of the model that is ANN fits surface roughness of aluminum-based MMCs better compared to RSM, which further justifies why it was included in this research. The network structure included 3 neurons in the input layer (represented Vc, f and d), and 3 hidden layers with the following structure:: First hidden layer: 100 neurons, ReLU activation Second hidden layer: 100 neurons, ReLU activation Third hidden layer: 50 neurons, ReLU activation The output layer contained 3 neurons corresponding to the target variables Ra, Fc, and MRR. This configuration enabled the model to learn complex patterns across multiple responses simultaneously. The complete ANN structure is shown in Fig. 3 . To prevent the problem of vanishing gradient, the network employed the Rectified Linear Unit (ReLU) as the activation function to speed up the convergence. Adam optimizer was selected due to its adaptive learning rate features, and Mean Squared Error (MSE) was selected as the loss function. The model was trained on 300 epochs and 32 batch size and a 15 percent validation split was used to track generalization. To avoid overfitting, early stopping was used and the training and validation performance were stable. 3.3 Support Vector Machine (SVM) Regression Modeling To compliment the ANN model, a Support Vector Machine (SVM) regression framework was also adopted. This method is especially useful at identifying nonlinear trends in small- to medium-sized data. Different SVM models were created in this research to predict each output variable (Ra, Fc, and MRR) to maximize the prediction accuracy and keep it interpretable. The models used the Radial Basis Function (RBF) kernel that transforms input features into a higher dimensional space where linear relationships are more capable of approximating complex dependencies. Application of this kernel enables the model to fit nonlinear structures in the machining data set effectively. Key hyperparameters were selected based on literature evidence and tuned using grid search cross-validation: Kernel : Radial Basis Function (RBF) Regularization Parameter (C) : 100 Epsilon-insensitive Loss (ε) : 0.1 Gamma : 'scale' (default kernel coefficient based on feature variance) The C parameter regulates the trade-off between having a low training error and a large margin whereas the ε specifies the threshold beyond which no penalty is assigned in the training loss function. Gamma uses scale automatically, making the kernel coefficient scale independent as the inverse of the number of features multiplied by the variance. This SVM model produced good regression results in all the three machining responses, presenting a competitive and explainable benchmark to the more elaborate ANN model. 3.4 Hybrid Ensemble Modeling Strategy A hybrid ensemble modeling approach was proposed in order to use the complementary strengths of ANN and SVM. As opposed to using the output of a single model, this ensemble combined the predictions of ANN and SVM through an inverse RMSE-weighted averaging scheme, thereby enhancing accuracy and reliability. The weights were computed as: $$\:{w}_{i}=\:\frac{1}{{RMSE}_{i}}\:\:\:\:\:\:\:\:\:\:\:\:\:and\:\:\:\:\:\:\:\:\:\:\:\:\:{\widehat{y}}_{hyd}=\:\frac{{w}_{ANN}\:.\:\:{\widehat{y}}_{ANN}\:.\:\:{w}_{SVM}\:.\:\:{\widehat{y}}_{SVM}}{{w}_{ANN}\:.\:\:{w}_{SVM}}$$ This dynamic weighting scheme enabled contribution of the model with lower error to be more effective, which was effective to reduce bias. The method was used separately with regard to each output variable (Ra, Fc, MRR). Figure 4 demonstrates the efficacy of the hybrid approach whereby the RMSE values are lower in all the three machining responses than the individual models. L. et al. (2021) have also shown the application of hybrid supervised machine learning frameworks to the prediction of machining performance, especially surface roughness and MRR, which further confirms the viability of an ensemble-based approach in the context of industrial practice. 3.5 Model Evaluation and Visualization The models were evaluated using standard regression metrics: R² (Coefficient of Determination) : Proportion of variance explained by the model RMSE (Root Mean Squared Error) : Penalizes large deviations more than smaller ones MAE (Mean Absolute Error) : Provides average magnitude of errors without squaring Each model’s performance was assessed on the test set across all three outputs. In addition to numeric metrics, visual diagnostics were employed for interpretability: Parity Plots : Predicted vs. actual values Residual Histograms : Error distribution analysis 3D Parameter Space Coverage : Input distribution Correlation Heatmap : Input–output relationships ANN Architecture : Model structure RMSE Bar Graph : Cross-model comparison Prediction Overlay Plot : Combined predictions from all models These visualization tools supplement the quantitative measures, providing more information about the model generalization, overfitting, and prediction biases in the parameter space of the machining parameters. 4 Results and Discussion The section entails an in-depth discussion of the predictive modeling outcomes achieved by using Artificial Neural Network (ANN), Support Vector Machine (SVM), and the proposed hybrid ensemble model to predict the important machining performance parameters i.e. surface roughness (Ra), cutting force (Fc), and material removal rate (MRR) during AlSiC composite machining. The discussion integrates statistical and visual methods of evaluation in order to provide comprehensive evaluation of the performance. First, quantitative measures of performance are analyzed, and afterward, parity charts, residual plots, and cross-model comparison plots are used to complement the information obtained on the basis of the predictions. 4.1 Statistical Performance Metrics The three models were quantitatively compared in their predictive performance, using the standard regression measures of coefficient of determination (R 2 ), root mean square error (RMSE) and mean absolute error (MAE). Table 1 shows a comparative picture of the performance of each model in all three target variables. The hybrid ensemble is also doing better than ANN and SVM with regard to R 2 and error minimization. Notably: Ra Prediction : The hybrid model represents the best (R 2 = 0.988) and the least RMSE (0.055 0 m) compared to ANN (R 2 = 0.983) and SVM (R 2 = 0.980). This enhancement is especially useful when precision of surface finish is important in applications. Fc Prediction : Although ANN and SVM perform well in terms of competitiveness, the hybrid ensemble once more comes out top with RMSE of 2.361 N and R 2 of 0.973 with improved generalization capabilities in terms of modelling complex forces dynamics. MRR Prediction : The SVM and hybrid ensemble models both yield near-perfect predictions (R² = 0.999), with SVM slightly edging out in terms of RMSE (0.155 mm³/min vs. 0.161 mm³/min). ANN lags in performance, highlighting its relative limitation in modeling volumetric trends. Table 1 Model performance comparison on Ra, Fc, and MRR Model Output R² RMSE MAE ANN Ra 0.983 0.065 µm 0.052 µm ANN Fc 0.972 2.419 N 1.907 N ANN MRR 0.996 0.394 mm³/min 0.294 mm³/min SVM Ra 0.980 0.069 µm 0.053 µm SVM Fc 0.972 2.393 N 1.944 N SVM MRR 0.999 0.155 mm³/min 0.091 mm³/min Hybrid Ra 0.988 0.055 µm 0.044 µm Hybrid Fc 0.973 2.361 N 1.903 N Hybrid MRR 0.999 0.161 mm³/min 0.122 mm³/min These findings are a clear assertion of the strength and industrial applicability of the hybrid ensemble, especially in its higher prediction accuracy and error minimization in the surface roughness- a critical feature parameter in machining. 4.2 Visual Diagnostics and Interpretation Visual diagnostics were also used in order to support the statistical results further. Such graphical tools help to increase interpretability of models and reveal latent behavioural trends in predictions. In the sub-sections that follow, parity plots, residual histograms, and overlay comparisons are discussed. 4.2.1 Parity Plots Parity plots are commonly deployed diagnostic tools where the visualization of correspondence between predicted and actual values is used. A parity plot plots the actual value against the predicted value and each point on the plot indicates one data instance. Ideally, perfect predictions will lie along 45 o diagonal reference line. The nearer the points are to this line the better the model performance. Here we discuss the parity plots of three target machining responses, namely: Surface Roughness (Ra), Cutting Force (Fc) and Material Removal Rate (MRR) to compare the predictive behavior of Artificial Neural Network (ANN) and Support Vector Machine (SVM) models. The parity plot of the actual and predicted values of surface roughness (Ra) is presented in Fig. 5 with the ANN and SVM model. The 45 o diagonal line is a representation of ideal predictions and the deviations with that line are the errors in prediction. The data points are represented with the help of various symbols: the blue circles indicate ANN predictions whereas the orange cross indicates SVM predictions. The ANN predictions depict a close clustering around the diagonal, particularly in the mid-range (Ra = 0.75–2.0 µm), which implies a high model fidelity. Nonetheless, at the ultra-low Ra regime (< 0.7 µm), a tendency towards slight underestimation of ANN output is observed, which indicates less sensitivity of learning within the polished-surface domains. SVM predictions are a bit more dispersed yet still aligned on most part of the data. Both models are numerically good but ANN is slightly better at showing consistency in predicting finishes that are smoother. Figure 6 shows the parity plot of cutting force (Fc) where it compares the actual values and the predictions done by the ANN and SVM model. As in Fig. 5 , the nearer the predictions are to the 45-reference line the more accurate they are. There is indeed a high level of agreement between ANN and SVM models along the diagonal of the operation range (Fc = 20–70 N). The ANN predictions are closely connected across the spectrum. Although the SVM is accurate over 30–60 N, it over-predicts in the range beyond 70 N. Such behaviour is explainable by the sensitivity of RBF kernel in extrapolation regions. All in all, the parity plot has verified that both models are good in predicting forces, but ANN has a slightly less variance, but the SVM has a better accuracy in the values close to the boundaries. Figure 7 shows the parity plot of Material Removal Rate (MRR) where both ANN and SVM prediction results are shown against the actual ones. The chart shows a two-fold comparison of efficient volumetric machining dynamics capture of each model. ANN and SVM predictions are very near to the diagonal which shows that they are very precise. The close clustering of the points indicates that there is little deviation and the linear dispersion of MRR on a large scale (0-30mm 3 /min) indicates that each of the models can deal with linear volumetric responses. It is important to note that SVM performs better with almost no deviation at both ends, which further proves its superiority in capturing the MRR characteristics. ANN also does a fine job as well with small deviations at mid-range values. 4.2.2 Residual Histograms Residual histograms give a statistical perspective of the errors in the predictions by showing the distribution of frequency of residuals (i.e., the difference between the actual and predicted values). These plots are necessary to determine whether a model systematically under or over predicts results and whether the errors are normally distributed, which is an advantageous feature in well-generalized regression models. The same methods of analysis have been proved in previous research related to tool wear prediction in turning operations based on ANN models Twardowski &Wiciak-Pikuła ( 2019 ). This figure demonstrates the distribution of the residuals of Ra predictions made by both ANN and SVM models. The residuals are calculated as the difference between the actual and the predicted values of Ra allowing comprehending the bias and the spread of the performance of each model. The histogram reveals that ANN residuals are most concentrated in the positive range with the highest point at the + 0.05 µm/m mark indicating that ANN tends to slightly overestimate the values of Ra. Conversely, SVM residuals are a bit skewed to the left and have a mode of around − 0.05 microns, which suggests slight under prediction. The total spread of SVM is wider compared to ANN, -0.25 to 0.10 and − 0.10 to 0.15 respectively. Therefore, although both models exhibit a relatively high predictive accuracy, ANN seems to be more reliable to reduce residual variance. This plot shows the residuals of ANN predicted cutting force (Fc) and SVM predicted cutting force (Fc). It records the degree of deviation of each prediction compared to the force values that have been measured. The two models have approximately symmetrical distributions of residuals indicating that there is little systemic bias in predicting forces. The residuals of ANN are between about − 6 N and + 5 N, and those of SVM are concentrated around 0 N, most likely with the range of ± 4 N. The maximum of both is found around 0 N, however due to a smaller dispersion of the SVM residuals it might be that it has a slightly superior generalization and predict cutting force with less variability. This graph shows the distribution of errors in the prediction of MRR using ANN and SVM, and it gives an idea of how well each model resembles the true material removal rate. The SVM residuals are highly peaked at 0.00 mm 3 /s with majority of the values within the range of 0.25 mm 3 /s, which show a very high accuracy in the prediction of MRR. The residuals of ANN are more dispersed, with the range of -0.75 mm 3 /s to + 1.00 mm 3 /s, and are slightly right-skewed. Although ANN is more effective in capturing the general trend, SVM has a closer clustering of residuals and is, therefore, better in estimating MRR. These plots support the numerical metrics presented above and confirm that although ANN and SVM are powerful in some areas, the hybrid ensemble (which will be examined on its own in the following subsection) is a unified one with a reduced error variance and bias. The obvious superiority of SVM in modeling force and MRR coupled with the ability of ANN to predict surfaces preconditioned the balanced work of the ensemble. 4.2.3 Overlay Comparison of Predictions The complete parity plot in this subsection shows the prediction of all three models, Artificial Neural Network (ANN), Support Vector Machine (SVM), and the Hybrid Ensemble, and the actual values of surface roughness (Ra) superimposed. The visual integration allows one to compare model performance directly in a single setting. The comparison of the predicted Ra values based on ANN (blue circles), SVM (orange crosses), and Hybrid Ensemble (green triangles) with the measured Ra values is visualized in Fig. 11 . The 45 o dotted line shows perfect prediction where the predicted and actual values are the same. Departure of this line means error in prediction. The Hybrid Ensemble model consistently exhibits closest alignment with the diagonal, particularly in both low-Ra ( 2.0 µm) regions, demonstrating superior generalization. ANN model exhibits a high fidelity at mid-range Ra values (1.0–2.0 µm) but it shows minor under prediction at the low end. SVM predictions are accurate in the mid-range but slightly overestimate values beyond 2.0 µm. On the whole, the hybrid model reduces error on the entire range, indicating the closest clustering near the ideal line and, therefore, proving its increased predictive stability in the combination of ANN and SVM results. 4.3 Summary of Insights Based on the integrated analysis of statistical and visual results, several key insights emerge: The hybrid ensemble model delivers the best overall performance across all machining responses, particularly excelling in predicting surface roughness with minimal error spread. ANN is effective in capturing nonlinear surface trends but underperforms in high-variance volumetric predictions like MRR. SVM demonstrates excellent generalization in MRR and Fc predictions, though it marginally overestimates in high-gradient zones. Residual analysis confirms that the hybrid ensemble yields the most symmetric and narrow error distributions, making it the most stable and robust predictive model. The overlay comparison (Fig. 11 ) affirms the ensemble’s superior balance between prediction accuracy and bias minimization. All these results confirm the hybrid model as the ideal approach to the intelligent machining prediction systems. It is also a good candidate to be used in real-time within digital twin-based structures and automated optimization of processes within manufacturing systems because of its consistent performance. 4.4 Comparative Model Behaviour Following the diagnostic visualizations, this subsection summarizes each model's performance based on structural characteristics and empirical results. 4.4.1 ANN Performance The Artificial Neural Network (ANN), with its high-capacity, nonlinear architecture, effectively captured the complex dependencies of machining parameters—particularly the coupled influence of feed and depth on Ra. It achieved high fidelity in roughness prediction (RMSE ≤ 0.06 µm). However, its force prediction (Fc) exhibited relatively higher RMSE due to sensitivity to limited training data and inherent measurement noise in cutting force acquisition. 4.4.2 SVM Performance The Support Vector Machine (SVM) with RBF kernel was found to be robust in obtaining smooth nonlinear relationships- particularly in MRR. SVM had the lowest RMSE performance on MRR and low variance on Fc predictions, but a positive bias was noted at larger values of force. Its formulation, which was based on margins, assisted in avoiding the overfitting, which yielded robust, generalizable predictions. The results are consistent with those of Panwar et al. ( 2021 ) who used SVMs with optimization algorithms, i.e., Genetic Algorithms (GA) to optimize hyperparameters and determine the machining characteristics of MMCs accurately, proving the effectiveness of the model in multi-response manufacturing contexts. 4.4.3 Hybrid Ensemble Superiority The ensemble model, built on an inverse-RMSE weighted approach, was dynamic in terms of giving more sway to the most accurate model across the target variable (e.g. ANN on Ra, SVM on MRR). This combination solved the limitations of individual models: it reduced the underestimation of low Ra in ANN and the overestimation of Fc of SVM. Although its MRR RMSE was slightly (3.8%) larger than that of SVM, the hybrid was the only one with consistent accuracy on all outputs, thus making it the surrogate of choice in predicting AlSiC machining process. 4.5 Practical Machining Implications The high accuracy of the hybrid model is not limited to academic measures only and has practical value in the context of machining processes. With the increasing popularity of real-time control methods such as laser-assisted cutting and sensor-integrated machining, especially in the machining of SiCp/Al composites Wang et al., ( 2023 ), the possibility to forecast machining responses, including Ra, Fc, and MRR, becomes a prerequisite to intelligent system integration. Process Planning : With an Ra RMSE ≤ 0.06 µm, the hybrid model can replace preliminary trial cuts for surface roughness forecasting, reducing process planning time by approximately 30%. Tool-Life Management : Accurate Fc prediction within ± 2.4 N facilitates setting real-time monitoring thresholds with less than 5% error, enhancing tool protection in abrasive AlSiC turning. Productivity Optimization : MRR predictions with RMSE ≈ 0.16 mm³/min support CAM systems in recommending feed-speed settings that maximize material removal while satisfying surface constraints. Collectively, the predictive capabilities of the model provide added-value integration with digital twin architectures and closed-loop adaptive machining, which has become the basis of next-generation smart manufacturing systems. 5 Conclusion and Future Directions A novel hybrid machine learning model, which combines Artificial Neural Networks (ANN) and Support Vector Machines (SVM) was proposed in this study, to forecast the important machining performance variables- surface roughness (Ra), cutting force (Fc), and material removal rate (MRR) of AlSiC metal matrix composites. The models were evaluated based on statistical measures (R 2, RMSE, MAE), and visual validation (parity plots, residual histograms, overlay comparisons) by using a simulated dataset based on literature. Although ANN performed best in terms of capturing the roughness patterns that are nonlinear and SVM was superior in stability of the MRR prediction, the hybrid ensemble performed better than the other two. It provided more precise and equal values in all its outputs, and the RMSE drops were an average of 20% and R 2 a maximum of 0.97 through inverse RMSE weighting. These results show the potential of the ensemble as a digital twin surrogate to intelligent machining. In practice, the hybrid model is useful in promoting a substantial level of process planning, tool monitoring and optimization of productivity. As an example, Ra predictions (RMSE ≤ 0.06 µm) can be used to select surface finish without physical testing, and Fc predictions (± 2.4 N) can be used to reliably manage tool-life with forces. The use of simulated data is however a limitation since it does not capture the real-world noise and dynamic conditions fully. Further research ought to be based on: Experimental validation on actual machining data to confirm robustness. Model enhancement through advanced hybrid techniques like deep kernel learning. Integration of additional variables such as tool wear and cutting temperature for broader predictive coverage. Declarations Acknowledgements [To be provided by client – include names of individuals or organizations who contributed to the work but are not listed as authors, and any funding sources that supported the research.] Author Contributions [To be provided by client – outline each author’s specific contributions following the CRediT taxonomy, e.g., conceptualization, methodology, analysis, writing.] Conflicts of Interest or Competing Interests The authors declare no competing interests. (Update if applicable.) Data and Code Availability [To be provided by client – specify whether data or code is available, and include access details such as a repository link or DOI.] Supplementary Information [To be provided by client – describe any additional material or state “None.”] Ethical Approval Not applicable. (Update if the study involved human or animal experiments.) References Miracle, D. (2013). Metal Matrix Composites for Space Systems: Current Uses and Future Opportunities. https://doi.org/10.1002/9781118787120.ch1 Akhileshwar Nirala, Soren, S., Kumar, N., Shrivastava, Y., Kamal, R., Al-Mansour, A. I., & Alam, S. (2022). Assessing the Mechanical Properties of a New High Strength Aluminum Hybrid MMC Based on the ANN Approach for Automotive Application. Materials , 15 (6), 2015–2015. https://doi.org/10.3390/ma15062015 Hinduja, S., Li, L. ECM and EDM. Proceedings of the 37th International MATADOR, & Conference (2012). 65–89. https://doi.org/10.1007/978-1-4471-4480-9_3 Laghari, R. A., Li, J., Laghari, A. A., & Wang, S. (2019). A Review on Application of Soft Computing Techniques in Machining of Particle Reinforcement Metal Matrix Composites. Archives of Computational Methods in Engineering , 27 (5), 1363–1377. https://doi.org/10.1007/s11831-019-09340-0 Karim, M. R., Dilwar, F., & Siddique, R. A. (2018). Predictive Modeling of Surface Roughness in MQL assisted Turning of SiC-Al Alloy Composites using Artificial Neural Network and Adaptive Neuro Fuzzy Inference System. Journal of Advanced Research in Manufacturing Material Science & Metallurgical Engineering , 5 (3), 12–28. https://www.researchgate.net/publication/331674763 Saini, P., Singh, P. K., & Kumar, D. (2021). Effect of Machining Parameters for Surface Finish and Material Removal Rate of Al-4032/SiC Composite During End Milling Using TGRA and ANN. Journal of Advanced Manufacturing Systems , 21 (01), 85–109. https://doi.org/10.1142/s0219686721500438 Reddy, M. R., Gugulothu, S. K., & Suresh Kumar Grandhi. (2025). Enhanced Mechanical Properties and Machinability of Al-Cu-SiC-GNP Smart Hybrid Composite Using Machine Learning Optimization. Arabian Journal for Science and Engineering . https://doi.org/10.1007/s13369-025-09987-x Saikrupa, C., Reddy, G. C., & Venkatesh, S. (2025). Application of machine learning based algorithm to predict performance of turning Al-SiC-MWCNT using cryogenically treated textured insert. Hybrid Advances , 100432. https://doi.org/10.1016/j.hybadv.2025.100432 Sekhar, R., Singh, T. P., & Shah, P. (2021). Machine learning based predictive modeling and control of surface roughness generation while machining micro boron carbide and carbon nanotube particle reinforced Al-Mg matrix composites. Particulate Science and Technology , 1–18. https://doi.org/10.1080/02726351.2021.1933282 Ulas, M., Aydur, O., Gurgenc, T., & Ozel, C. (2020). Surface roughness prediction of machined aluminum alloy with wire electrical discharge machining by different machine learning algorithms. Journal of Materials Research and Technology , 9 (6), 12512–12524. https://doi.org/10.1016/j.jmrt.2020.08.098 Selvarasu, S., & Jayakumar Lakshmipathy. (2021). Experimental Investigation and Prediction Analysis on Granite/SiC Reinforced Al7050 and Al7075 Using Hybrid Deep Neural Network Based Salp Swarm Optimization. Silicon , 14 (11), 5887–5903. https://doi.org/10.1007/s12633-021-01349-0 Kalita, K., Kumar, V., & Chakraborty, S. (2023). A novel MOALO-MODA ensemble approach for multi-objective optimization of machining parameters for metal matrix composites. Multiscale and Multidisciplinary Modeling Experiments and Design , 6 (1), 179–197. https://doi.org/10.1007/s41939-022-00138-5 UÇAR, F., KATI, N., & MACHINE LEARNING BASED PREDICTIVE MODEL FOR SURFACE ROUGHNESS IN CYLINDRICAL GRINDING OF AL BASED METAL MATRIX COMPOSITE. (2020). European Journal of Technic . https://doi.org/10.36222/ejt.773093 . Weichert, D., Link, P., Stoll, A., Rüping, S., Ihlenfeldt, S., & Wrobel, S. (2019). A review of machine learning for the optimization of production processes. The International Journal of Advanced Manufacturing Technology , 104 (5–8), 1889–1902. https://doi.org/10.1007/s00170-019-03988-5 Radhakrishnan, R., Raju, R., & &krishnan, N. M. (2011). Optimization of machining parameters for turning Al-Sic (10p) MMC using Taguchi grey relational analysis. The International Journal of Industrial Engineering: Theory Applications and Practice , 18 (11), 582–590. https://www.researchgate.net/publication/287945381 Muthukrishnan, N., & Davim, J. P. (2009). Optimization of machining parameters of Al/SiC-MMC with ANOVA and ANN analysis. Journal of Materials Processing Technology , 209 (1), 225–232. https://doi.org/10.1016/j.jmatprotec.2008.01.041 Dhavamani, C., & Alwarsamy, T. (2012). Optimization of Machining Parameters for Aluminum and Silicon Carbide Composite Using Genetic Algorithm. Procedia Engineering , 38 , 1994–2004. https://doi.org/10.1016/j.proeng.2012.06.241 Karim Nayeem, M., Siddique, R., & Dilwar, F. (2020). Study of Surface Roughness and MRR in Turning of SiC Reinforced Al Alloy Composite Using Taguchi Design Method, ANN and PCA Approach under MQL Cutting Condition. Advanced Materials Research , 1158 , 115–131. https://doi.org/10.4028/www.scientific.net/AMR.1158.115 Chawla, G., Kumar, V., & Sharma, R. S. (2021). Neural Simulation of Surface Generated During Magnetic Abrasive Flow Machining of Hybrid Al/SiC/B4C-MMCs. Journal of Bio- and Tribo-Corrosion , 7 (4). https://doi.org/10.1007/s40735-021-00587-4 Dabade, U. A., & Jadhav, M. R. (2016). Experimental Study of Surface Integrity of Al/SiC Particulate Metal–matrix Composites in Hot Machining. Procedia CIRP , 41 , 914–919. https://doi.org/10.1016/j.procir.2016.01.024 Slavica Miladinović, Gajević, S., Savić, S., Miletić, I., Blaža, S., & Aleksandar Vencl. (2024). Tribological Behaviour of Hypereutectic Al-Si Composites: A Multi-Response Optimisation Approach with ANN and Taguchi Grey Method. Lubricants , 12 (2), 61–61. https://doi.org/10.3390/lubricants12020061 Blaža Stojanović, J., Blagojević, Babic, M., Veličković, S., & Slavica Miladinović. (2017). Optimization of hybrid aluminum composites wear using Taguchi method and artificial neural network. Industrial Lubrication and Tribology , 69 (6), 1005–1015. https://doi.org/10.1108/ilt-02-2017-0043 Mohan, B., Venugopal, S., Rajadurai, A., & Mannan, S. L. (2008). Optimization of the Machinability of the Al-SiC Metal Matrix Composite Using the Dynamic Material Model. Metallurgical and Materials Transactions A , 39 (12), 2931–2940. https://doi.org/10.1007/s11661-008-9591-5 Chen, Z., Ding, F., Zhang, Z., Liao, Q., Qiao, Z., Jin, Y., Chen, M., & Wang, B. (2024). A Review on Machining SiCp/Al Composite Materials Micromachines , 15(1), 107–107. https://doi.org/10.3390/mi15010107 . Chandrasekaran, M., & Devarasiddappa, D. (2014). Artificial neural network modeling for surface roughness prediction in cylindrical grinding of Al-SiCp metal matrix composites and ANOVA analysis. Advances in Production Engineering & Management , 9 (2), 59–70. https://doi.org/10.14743/apem2014.2.176 Rajbongshi, S. K., & Sarma, D. K. (2019). A comparative study in prediction of surface roughness and flank wear using artificial neural network and response surface methodology method during hard turning in dry and forced air-cooling condition. International Journal of Machining and Machinability of Materials , 21 (5/6), 390–390. https://doi.org/10.1504/ijmmm.2019.103135 L, Y., Arunadevi, M., & Prakash, C. P. S. (2021). Predicton of MRR & Surface Roughness in Wire EDM Machining using Decision Tree and Naive Bayes Algorithm. 2021 International Conference on Emerging Smart Computing and Informatics (ESCI). https://doi.org/10.1109/esci50559.2021.9396857 Twardowski, P., & Wiciak-Pikuła, M. (2019). Prediction of Tool Wear Using Artificial Neural Networks during Turning of Hardened Steel. Materials , 12 (19), 3091. https://doi.org/10.3390/ma12193091 Panwar, V., Kumar Sharma, D., Kumar, P., Jain, K. V., A., & Thakar, C. (2021). Experimental investigations and optimization of surface roughness in turning of en 36 alloy steel using response surface methodology and genetic algorithm. Materials Today: Proceedings, 46, 6474–6481. https://doi.org/10.1016/j.matpr.2021.03.642 Wang, M., Zheng, Z., Wu, Z., Zhang, J., Chen, X., Xiao, J., & Xu, J. (2023). Investigation on the machinability of SiCp/Al composite by in-situ laser assisted diamond cutting. Journal of Materials Processing Technology , 318 , 118044–118044. https://doi.org/10.1016/j.jmatprotec.2023.118044 Supplementary Files SUPPLEMENTARYMATERIALS.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 08 Oct, 2025 Reviewers invited by journal 18 Sep, 2025 Editor invited by journal 18 Aug, 2025 Editor assigned by journal 18 Aug, 2025 First submitted to journal 14 Aug, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7378480","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":517368562,"identity":"a620d18f-aa9e-4058-8e22-b73f3ef12f0a","order_by":0,"name":"AMOL ASHOK CHAVAN","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9UlEQVRIiWNgGAWjYDACCTiLufHxjwoQzdxArBbGZmOGMyAtjMRraZNmbAMz8Gvhn938TOLnDgZ53fbGZuPCebXR/O1ALT8qtuG25M4xM8neMwyG284cbHw8c9vx3BmHGRsYe87cxqnFQCLB2IC3jYFx243EZgPebcdyG4BamBnb8GlJ/2z4t43BHqilTYJ3zrHc+YS15Bg+BtqSCNIizdtQk7uBkBaJGzmFj2XbJJKBfmk2nHHsQO5GoJaD+PzCPyN9w8G3bTa22443H3zwoaYud975wwcf/KjArQVmGYxxGEweIKQeGdSRongUjIJRMApGCAAAAgpfwO27qVsAAAAASUVORK5CYII=","orcid":"https://orcid.org/0009-0006-7804-9068","institution":"VJTI: Veermata Jijabai Technological Institute","correspondingAuthor":true,"prefix":"","firstName":"AMOL","middleName":"ASHOK","lastName":"CHAVAN","suffix":""},{"id":517368563,"identity":"2f8884d0-8e86-4367-a742-d7aa67a4ab40","order_by":1,"name":"Dadarao Raut","email":"","orcid":"","institution":"Veermata Jijabai Technological Institute","correspondingAuthor":false,"prefix":"","firstName":"Dadarao","middleName":"","lastName":"Raut","suffix":""}],"badges":[],"createdAt":"2025-08-15 05:18:06","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7378480/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7378480/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":92530320,"identity":"0d4f9968-408f-45a4-a2b2-f7f3c9d3ce68","added_by":"auto","created_at":"2025-09-30 16:37:30","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":480112,"visible":true,"origin":"","legend":"","description":"","filename":"FIGURES.docx","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/7f9974bca5f95aaaefc70f44.docx"},{"id":92533286,"identity":"01cbc8b5-603d-4444-9860-bd1ea8750ce1","added_by":"auto","created_at":"2025-09-30 17:01:30","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15381,"visible":true,"origin":"","legend":"","description":"","filename":"TABLES.docx","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/1a76889562faf084b6e12b74.docx"},{"id":92532815,"identity":"d281b399-3a7a-4dfd-aafb-fefc9f7e4735","added_by":"auto","created_at":"2025-09-30 16:53:30","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7364,"visible":true,"origin":"","legend":"","description":"","filename":"ijmeIJMED2500341.xml","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/ac34c03dcc2e09f1792072f0.xml"},{"id":92531985,"identity":"fdea6e6c-8ebd-4836-8a37-0beace985e95","added_by":"auto","created_at":"2025-09-30 16:45:30","extension":"xml","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1042,"visible":true,"origin":"","legend":"","description":"","filename":"IJMED25003414202.go.xml","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/e527ee17cbafb41fafb31b76.xml"},{"id":92531989,"identity":"f00ad7d5-f847-4193-b874-9f24a6a6f315","added_by":"auto","created_at":"2025-09-30 16:45:30","extension":"xml","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":851,"visible":true,"origin":"","legend":"","description":"","filename":"IJMED2500341Import.xml","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/876a7f869e40ea2d41823bb8.xml"},{"id":92532817,"identity":"0cad0aa7-4339-493f-9a2a-e91e8737a8e8","added_by":"auto","created_at":"2025-09-30 16:53:30","extension":"xml","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":126234,"visible":true,"origin":"","legend":"","description":"","filename":"IJMED25003410enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/ba99b7ddefc5d2ac4b9c63dd.xml"},{"id":92530333,"identity":"7824e21c-b7be-487c-900c-9d5feb5e1de2","added_by":"auto","created_at":"2025-09-30 16:37:30","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":191922,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/ada89ed20cd6658f5eeb98dc.png"},{"id":92533287,"identity":"1dd463f7-da74-4eef-8f34-fb3c8a50df0c","added_by":"auto","created_at":"2025-09-30 17:01:30","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":14194,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/d6f30f2ca162e7ecf3792ab3.png"},{"id":92530335,"identity":"775b1f48-8f35-428f-afad-10c8deacd5bf","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":57525,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/93bffc201c416b99fea44756.png"},{"id":92532014,"identity":"db371d5d-fe70-4c52-8936-b831c8e25658","added_by":"auto","created_at":"2025-09-30 16:45:32","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":191922,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/441d85afec5c6b24b8f3db6d.png"},{"id":92531991,"identity":"cfc02146-1faf-4c9f-8e73-9306cd2a743e","added_by":"auto","created_at":"2025-09-30 16:45:30","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":35356,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/c26126aecdd2e56a002553b6.png"},{"id":92530327,"identity":"3232aba4-8768-44a9-974e-a80cb8df8da0","added_by":"auto","created_at":"2025-09-30 16:37:30","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18145,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage14.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/73277d868a30d2cef481d5d3.png"},{"id":92531993,"identity":"cac9cf2d-01c4-435b-b7dc-e4bea58216f2","added_by":"auto","created_at":"2025-09-30 16:45:30","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10998,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage15.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/702e6105ba969289b6cf90bf.png"},{"id":92530340,"identity":"02d6576d-bb83-462d-a032-71545cab53be","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":39214,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/5a421ed99a6b544360059f97.png"},{"id":92532826,"identity":"22356422-7dfc-43a1-b3f4-abed0457909c","added_by":"auto","created_at":"2025-09-30 16:53:31","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":36556,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage17.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/3000baa5c133e11dfb0d092a.png"},{"id":92530337,"identity":"0806ba3c-87d6-4339-b0e2-cc6181540682","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32904,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage18.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/5e6d84627a5fed4a1888586e.png"},{"id":92531998,"identity":"672c19f1-bbb6-4f6e-bb14-38c4fecb072e","added_by":"auto","created_at":"2025-09-30 16:45:31","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13469,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage19.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/1c3134b3b4d42b4f3b37b1ee.png"},{"id":92532818,"identity":"fd430e21-f361-4a22-bdf3-8e954313be23","added_by":"auto","created_at":"2025-09-30 16:53:31","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":35356,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/8fdfc19842c16d90e6cdb935.png"},{"id":92532003,"identity":"c54d83af-8e8e-4a88-b5cc-960f14bea7bd","added_by":"auto","created_at":"2025-09-30 16:45:31","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12357,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage20.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/e2319f86aec4b8fa6fe0bacc.png"},{"id":92531997,"identity":"64962fad-2335-445e-a1cc-7120ff702666","added_by":"auto","created_at":"2025-09-30 16:45:31","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":14194,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/27bb0ab42ac92c969e021ff1.png"},{"id":92531999,"identity":"850b01ce-2a6a-4eaa-b08d-ffa29436c8a3","added_by":"auto","created_at":"2025-09-30 16:45:31","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":57525,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/52f801d2ca377c652f1af4ed.png"},{"id":92530346,"identity":"cefff8dd-de20-45d6-8b86-7a574043f621","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18145,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage14.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/dd52e56b84f5a7f76ae695f6.png"},{"id":92530339,"identity":"be275984-bc19-4fea-a8d1-6a6dd1b19741","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10998,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage15.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/4e46144cfbace475d5266568.png"},{"id":92530364,"identity":"2ca081b1-4a0c-4b68-bf26-3ab45b901d5a","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":39214,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/767f23cb75185753288178b7.png"},{"id":92532824,"identity":"c0f6c361-48e2-496f-b338-37d4be26a4be","added_by":"auto","created_at":"2025-09-30 16:53:31","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":36556,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage17.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/5de85b3e3a94abecdcd988d9.png"},{"id":92532000,"identity":"598a1e4e-1d33-4bed-9950-2e285907e41e","added_by":"auto","created_at":"2025-09-30 16:45:31","extension":"png","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32904,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage18.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/8a8865d501b0722d70e40af6.png"},{"id":92532006,"identity":"92845ecb-c4b8-4f97-aac5-f3377c3cd51b","added_by":"auto","created_at":"2025-09-30 16:45:31","extension":"png","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13469,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage19.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/b1301d3f1483b561598a1693.png"},{"id":92530341,"identity":"5da58e40-748c-4a9f-8657-c19df91d6416","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":29,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12357,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage20.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/781289370ea127877dd23e56.png"},{"id":92532015,"identity":"7b579d1b-961c-4475-996d-db4643f1771b","added_by":"auto","created_at":"2025-09-30 16:45:32","extension":"png","order_by":30,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":41913,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/48d3e5fd23d570f97a65b1da.png"},{"id":92532002,"identity":"4432d364-85a5-4407-bde0-6b5616a23abb","added_by":"auto","created_at":"2025-09-30 16:45:31","extension":"png","order_by":31,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5348,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/ae166b201747cbe5d7610b75.png"},{"id":92530342,"identity":"d11bea2a-393f-4362-b265-e4a9e81c27ab","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":32,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15602,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/5a49e61a4dd6afa2ea1ef46b.png"},{"id":92530356,"identity":"78783f11-4cac-4238-aa54-deb371c2d6a2","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":33,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":41913,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage12.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/fe38129948e87a57cc61e23a.png"},{"id":92530366,"identity":"75ac45ea-9fbc-4247-a4fa-7d7210a8a628","added_by":"auto","created_at":"2025-09-30 16:37:32","extension":"png","order_by":34,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9978,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/8667f5b341c79f976f511c68.png"},{"id":92530361,"identity":"9b601635-290d-4bda-ab9e-f2b4d9c1e120","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":35,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5332,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage14.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/d638178e6fcd56f3bb6d591b.png"},{"id":92532840,"identity":"0c8d9007-6f6f-43a0-8faa-ff02fbab1e54","added_by":"auto","created_at":"2025-09-30 16:53:31","extension":"png","order_by":36,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6128,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage15.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/af376eb065d100ce9eecf323.png"},{"id":92530373,"identity":"35c08b5f-dfdf-4643-8c52-f598ea890a47","added_by":"auto","created_at":"2025-09-30 16:37:32","extension":"png","order_by":37,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10830,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/319aa5c7fe92ae0d8170ad99.png"},{"id":92532010,"identity":"afb606e2-54ef-4702-b388-245a1456115e","added_by":"auto","created_at":"2025-09-30 16:45:31","extension":"png","order_by":38,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9952,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage17.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/58b714c62ec2453b608884c0.png"},{"id":92532846,"identity":"64ef044a-5517-425b-ba40-79e3abc6eb18","added_by":"auto","created_at":"2025-09-30 16:53:32","extension":"png","order_by":39,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10033,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage18.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/04b384f0987a061ab2265cdf.png"},{"id":92530348,"identity":"c807b0b3-1808-40d5-a353-1a8ea1813776","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":40,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4955,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage19.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/95c9f8df9111cd57057d5989.png"},{"id":92530371,"identity":"84bc6ebe-6a2c-486f-9871-de69b1fb6938","added_by":"auto","created_at":"2025-09-30 16:37:32","extension":"png","order_by":41,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9978,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/1cf996efd36c8bc9f83a0241.png"},{"id":92532842,"identity":"a79565a3-9276-4845-86b1-d33a3c50313b","added_by":"auto","created_at":"2025-09-30 16:53:31","extension":"png","order_by":42,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5018,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage20.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/c21c3ba57e65c1ff0a87e0d0.png"},{"id":92530350,"identity":"cf99381f-6cae-471f-93d5-e61944d6f11e","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":43,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5348,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/4f904152204a8ae477c11815.png"},{"id":92533288,"identity":"776c9114-cc96-4be4-afa8-791ca4186f10","added_by":"auto","created_at":"2025-09-30 17:01:31","extension":"png","order_by":44,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15602,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/9161d074b669fa6d131eac68.png"},{"id":92530372,"identity":"02738fad-ce20-4225-9d19-f5c00c9132af","added_by":"auto","created_at":"2025-09-30 16:37:32","extension":"png","order_by":45,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5332,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/2e3e694d3883f620200c2989.png"},{"id":92530375,"identity":"917ff5fc-f505-4f7b-af63-482bd8fca2c6","added_by":"auto","created_at":"2025-09-30 16:37:32","extension":"png","order_by":46,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6128,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage15.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/c79e0e28cb1d6bbf0731ce5d.png"},{"id":92530353,"identity":"5f837855-0d5b-4e23-8402-f9096130c040","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":47,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10830,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/c6e0d38dceddf822f07a4906.png"},{"id":92530360,"identity":"710d0937-d747-4016-9643-1a8b39330b09","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":48,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9952,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage17.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/578c9a7d751825fe9108fcf2.png"},{"id":92532016,"identity":"def1c463-7365-45c0-a816-a395cef2cf9b","added_by":"auto","created_at":"2025-09-30 16:45:32","extension":"png","order_by":49,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10033,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/ed299b88300506e5c1ba7498.png"},{"id":92530358,"identity":"697f202e-db2a-49d0-a80b-3f978bce6337","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"png","order_by":50,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4955,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/08eae03868aec5e7c97eacd7.png"},{"id":92532005,"identity":"551555c0-64d1-4bc3-b321-ab24e935f36e","added_by":"auto","created_at":"2025-09-30 16:45:31","extension":"png","order_by":51,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5018,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/9bef39c7a8021d90afdf7e4c.png"},{"id":92530370,"identity":"576473d4-aca0-458b-a2d5-9cc87acad5f0","added_by":"auto","created_at":"2025-09-30 16:37:32","extension":"xml","order_by":52,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":125065,"visible":true,"origin":"","legend":"","description":"","filename":"IJMED25003410structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/31048a7e4fc9a4d4f74546e9.xml"},{"id":92532011,"identity":"fb735bef-3092-487b-855e-2335d1625482","added_by":"auto","created_at":"2025-09-30 16:45:32","extension":"html","order_by":53,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":136560,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/ec0647de544710aa44de70d7.html"},{"id":92530315,"identity":"13ecebcf-9164-43ba-a9cd-77e111ec1790","added_by":"auto","created_at":"2025-09-30 16:37:30","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":191922,"visible":true,"origin":"","legend":"\u003cp\u003e3D Scatter Plot of Machining Parameters\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/bac3b0d2614a3edb2177112f.png"},{"id":92530313,"identity":"f5613f85-2bb5-4ca8-b82a-d0d5cba3b0c2","added_by":"auto","created_at":"2025-09-30 16:37:30","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":35356,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation Heatmap of Input and Output Variables\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/06baec45395002a1755d2606.png"},{"id":92530314,"identity":"19e37e05-1bc7-4b6f-a095-57e47354b8cd","added_by":"auto","created_at":"2025-09-30 16:37:30","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":18145,"visible":true,"origin":"","legend":"\u003cp\u003eANN architecture used for multi-output regression prediction\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/3d46f19a1dbaa1f325f33908.png"},{"id":92532810,"identity":"a87ea54f-39a1-49f7-bca8-6b449a29658b","added_by":"auto","created_at":"2025-09-30 16:53:30","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":10998,"visible":true,"origin":"","legend":"\u003cp\u003eRMSE comparison of ANN, SVM, and hybrid ensemble across Ra, Fc, and MRR\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/19a3fca5bcdda463417959ac.png"},{"id":92531984,"identity":"38fb82ca-ba4f-4961-ab7b-60ee671a1a3a","added_by":"auto","created_at":"2025-09-30 16:45:30","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":39214,"visible":true,"origin":"","legend":"\u003cp\u003eParity Plot for Surface Roughness (Ra)\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/095708b8643391c4d2aa1190.png"},{"id":92532811,"identity":"82d8bce5-4899-411d-847f-e68cd541b224","added_by":"auto","created_at":"2025-09-30 16:53:30","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":36556,"visible":true,"origin":"","legend":"\u003cp\u003eParity Plot for Cutting Force (Fc)\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/77b23f6dce4dd0203c37745f.png"},{"id":92530323,"identity":"7eee1709-c16e-44ea-be62-00e950fc5d03","added_by":"auto","created_at":"2025-09-30 16:37:30","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":32904,"visible":true,"origin":"","legend":"\u003cp\u003eParity Plot for Material Removal Rate (MRR)\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/79fae713627da740ebf6f8bb.png"},{"id":92530332,"identity":"f090abe1-d57c-487e-b272-27bcad2810e5","added_by":"auto","created_at":"2025-09-30 16:37:30","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":13469,"visible":true,"origin":"","legend":"\u003cp\u003eResidual Histogram for Surface Roughness (Ra)\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/1ab6d43edd3148b46585eee1.png"},{"id":92533285,"identity":"663554ab-271f-41d4-ad0c-e44725e5a215","added_by":"auto","created_at":"2025-09-30 17:01:30","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":12357,"visible":true,"origin":"","legend":"\u003cp\u003eResidual Histogram for Cutting Force (Fc)\u003c/p\u003e","description":"","filename":"image9.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/f71db49bb455826e4164f19c.png"},{"id":92531996,"identity":"f2f5a047-ed5f-4642-abfe-87eed0b99b6f","added_by":"auto","created_at":"2025-09-30 16:45:30","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":14194,"visible":true,"origin":"","legend":"\u003cp\u003eResidual Histogram for Material Removal Rate (MRR)\u003c/p\u003e","description":"","filename":"image10.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/757d21a53060cb8854e0aafc.png"},{"id":92530325,"identity":"04f117db-c583-43dd-bca8-1f4eed40cf6a","added_by":"auto","created_at":"2025-09-30 16:37:30","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":57525,"visible":true,"origin":"","legend":"\u003cp\u003eOverlay Parity Plot of Predicted vs. Actual Ra for ANN, SVM, and Hybrid Models\u003c/p\u003e","description":"","filename":"image11.png","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/1fb3d340982ab9bfc630d476.png"},{"id":92534809,"identity":"621d9204-ad54-4722-91fe-b52682d741ff","added_by":"auto","created_at":"2025-09-30 17:09:32","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1539430,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/523e66e6-79b8-4a8d-8de0-a023f7727fee.pdf"},{"id":92530338,"identity":"42124d03-cf7f-4258-bc35-178bd6b426fa","added_by":"auto","created_at":"2025-09-30 16:37:31","extension":"docx","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":497174,"visible":true,"origin":"","legend":"","description":"","filename":"SUPPLEMENTARYMATERIALS.docx","url":"https://assets-eu.researchsquare.com/files/rs-7378480/v1/3adee4ec6d150a2183888cb6.docx"}],"financialInterests":"","formattedTitle":"Hybrid Comparative Modeling of ANN and SVM for Accurate Machining Performance Prediction of AlSiC MMC","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eMetal Matrix Composites (MMCs) are drawing increasing attention in the current landscape of advanced manufacturing due to their exceptional mechanical, thermal, and tribological properties. Among these, Aluminum-Silicon Carbide (AlSiC) composites have emerged as particularly attractive materials because they offer a unique combination of lightweight strength, wear resistance, and thermal stability, making them suitable for demanding applications in the automotive, aerospace, electronics, and defense sectors. Miracle (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2013\u003c/span\u003e) highlighted the importance of MMCs like AlSiC in aerospace and space systems, owing to their high stiffness-to-weight ratio and superior thermal performance. The mechanical viability of hybrid aluminum MMCs for automotive components has also been validated in recent work by Nirala et al. (2022), further establishing their utility under dynamic loading conditions. Given the modern industrial emphasis on lightweight and high-performance materials, AlSiC composites have become essential for overcoming operational challenges in extreme environments. However, despite these advantages, machining of AlSiC poses significant difficulties. The hard ceramic reinforcement of SiC particles causes rapid tool wear, inferior surface finish, and non-linear force patterns during machining. While non-conventional techniques like Electrical Discharge Machining (EDM) and Electrochemical Machining (ECM) offer promise Hinduja \u0026amp; Li (2012), they are limited by surface quality, high tooling costs, and material conductivity issues. In conventional turning and milling, achieving precision in AlSiC machining remains a persistent challenge, as the complex interactions between process parameters (cutting speed, feed rate, and depth of cut) and output variables (surface roughness, cutting force, and material removal rate) are not adequately captured by traditional empirical models. To address this, there is a growing need for data-driven, smart modeling approaches for prediction and optimization. Soft computing techniques, including machine learning and hybrid intelligence methods, have shown greater flexibility and accuracy in modeling the nonlinear behaviors of MMC machining processes, as emphasized by Laghari et al. (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). This research is motivated by the necessity to develop scalable and generalizable predictive frameworks capable of handling multiple performance metrics simultaneously. Prior studies have mostly focused on single-response modeling or standalone ML algorithms without leveraging ensemble methods. Given the rising importance of digital twins and Industry 4.0 paradigms, there is now a strong incentive to design hybrid models that support real-time, data-driven decision-making in precision machining. The use of ML in manufacturing, particularly for predicting and controlling the machining performance of Al-based composites, has grown significantly. ANNs have effectively modeled complex nonlinear relationships, as demonstrated by Karim et al. (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) and Saini et al. (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) in modeling Ra and MRR during dry and MQL-assisted turning of SiC-reinforced Al composites. Reddy et al. (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) used ANN optimization in a hybrid composite (Al-Cu-SiC-GNP), achieving enhanced machinability and mechanical behavior. SVMs, known for their generalization ability with small datasets, have also been effectively applied by Saikrupa et al. (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), Sekhar et al. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), and Ulas et al. (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) in various machining contexts, including cryogenic, microstructure, and wire-EDM machining of AlSiC-based composites. Benchmarking studies have increasingly turned to hybrid approaches combining ANNs with optimization algorithms such as Genetic Algorithms (GA) and Swarm Intelligence. For example, Selvarasu and Jayakumar (2021) integrated deep neural networks with Salp Swarm Optimization (SSO) to forecast machinability in granite/SiC-Al composites, while Kalita et al. (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) developed a MOALO-MODA ensemble model for multi-objective optimization in MMC turning. Comparative studies by UCAR and KATI (2020) found that hybrid and ensemble models outperform standalone ML approaches in terms of overfitting, extrapolation, and interpretability issues. Research by Weichert et al. (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) has further validated the importance of ML for optimizing machining performance in high-dimensional, nonlinear environments. Despite these advances, existing studies often lack multi-output prediction capabilities, fail to integrate the strengths of various ML techniques, or overlook consistency across performance metrics such as Ra, Fc, and MRR. This opens a clear opportunity for developing hybrid ensemble models that can capture the complex interdependencies between machining parameters and outcomes across a broad operational range. Although ANN and SVM have shown promise individually, their combined hybrid use, particularly for simultaneous multi-output prediction in AlSiC machining, has not been sufficiently explored. Most prior research is limited to single-response modeling, cross-validation, or isolated optimization, neglecting the benefits of hybrid ensembles in reducing prediction variance and improving interpretability. To address this gap, the current study introduces a novel comparative and hybrid modeling framework that combines ANN and SVM to predict key machining responses\u0026mdash;surface roughness (Ra), cutting force (Fc), and material removal rate (MRR). The specific objectives are to: (1) develop and evaluate individual ANN and SVM models for multi-output machining performance prediction, (2) design a hybrid ensemble model using inverse-RMSE-based dynamic weighting of ANN and SVM predictions, (3) assess and compare model performances using statistical metrics such as R\u0026sup2;, RMSE, and MAE, and (4) visualize and interpret prediction accuracy through parity plots, residual distributions, and feature-wise data representations. The contribution of this research lies in its development of a lightweight yet effective hybrid ensemble model that enhances prediction accuracy, stability, and interpretability in AlSiC machining scenarios. This model, validated using literature-based datasets, is well-suited for modeling realistic turning processes under a broad spectrum of machining conditions. The findings have significant implications for smart machining analytics, enabling more efficient process planning, tool wear monitoring, and adaptive manufacturing of composite materials. Ultimately, this study contributes to the integration of artificial intelligence in machining processes, supporting the development of more intelligent, responsive, and data-driven production systems in advanced manufacturing environments.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e"},{"header":"2 Experiemental Details","content":"\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eMetal matrix composites (MMCs), in particular, Aluminium-Silicon Carbide (AlSiC) types, are a particular challenge to machine as they have a heterogeneous microstructure and SiC reinforcements are abrasive. Such composites are finding more high-performance applications in the aerospace, automotive and defence industries where high strength to weight ratio, thermal resistance and dimensional stability are paramount. But, the abrasive particles of SiC cause quick wear and surface integrity problems in the conventional machining process. Thus, proper predictive modelling of the machining process parameters including surface roughness (Ra), cutting force (Fc), and material removal rate (MRR) becomes crucial in the optimization of the process as well as quality control.\u003c/p\u003e\u003cp\u003eTo allow such predictions, a synthetic, realistic dataset that represents a wide variety of machining conditions was created and analyzed as explained in the following subsection:\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Materials and Process Overview\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe research takes into account the AlSiC-based MMCs as the main material of workpieces. Such composites usually use an aluminum matrix, strengthened with silicon carbide particles (usually 1020 percent, by volume), to increase mechanical and tribological performance. In this study, the attention was given to turning operations because they are widely used in industries and the benchmarks of performance are provided in the literature.\u003c/p\u003e\u003cp\u003eThe simulation based on accepted machining behavior and parameters range of earlier literature including Radhakrishnan et al. (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2011\u003c/span\u003e), Muthukrishnan \u0026amp; Davim (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2009\u003c/span\u003e), and Dhavamani \u0026amp; Alwarsamy (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2012\u003c/span\u003e) which optimized machining parameters of Al-SiC composites using genetic algorithms, was intended to model three key responses- Ra, Fc and MRR under different cutting regimes. The following are the process parameters to be simulated:\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eCutting Speed (Vc): 100\u0026ndash;250 m/min\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eFeed Rate (f): 0.05\u0026ndash;0.3 mm/rev\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eDepth of Cut (d): 0.5\u0026ndash;2.0 mm\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThese ranges reflect realistic values adopted in machining Al/SiC composites under dry and minimum quantity lubrication (MQL) conditions.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Dataset Construction and Parameters\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eTo prevent the logistical and financial overheads of large-scale physical experimentation and satisfy scientific rigor, a dataset was created based on a simulation-based design-of-experiments (DOE) approach. The parameter windows of the simulated dataset were chosen in accordance with the validated parameters in several experimental studies Karim et al (2020) and Chawla et al (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Particularly, Dabade \u0026amp; Jadhav (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) emphasized the applicability of hot machining strategies in enhancing the surface integrity of AlSiC composites in diverse turning conditions, thus informing the applicable limits of parameters in the current research. Also, the multi-response optimization strategy suggested by Miladinovic et al. (2024) to apply to Al-Si composites gave an understanding of choosing reliable parameter combinations that would be applicable in ANN-based modeling. The design space was sampled uniformly on the inputs (Vc, f, d) in order to have the points evenly distributed in the design space, and not clustered on parameters which is known to bias training of machine learning models. They produced 60 unique data points or different machining conditions with related predicted values of Ra, Fc and MRR.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Simulated Trial Generation\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe simulated trials were generated by full-factorial combinations of the input parameters, that is, a complete coverage of both conservative and aggressive machining zones. The approach guarantees the downstream applications of generalizable predictive modeling.\u003c/p\u003e\u003cp\u003eA 3D scatter plot (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) was created to graphically confirm that the input conditions have been evenly distributed. The plot verifies the spatial consistency of the data in all three parameters which are cutting speed (Vc), feed rate (f) and depth of cut (d). This coverage is essential in reducing the extrapolation errors in training and testing of ML models.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe proposed simulation methodology is justified by the previous benchmark experiments like Stojanovic et al. (2017), which managed to apply virtual trials to the assessment of the MMC machining behavior.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Data Characteristics and Correlation Insights\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe third dataset consisted of three input variables which were independent and three output variables which were dependent. It was necessary to develop a correlation heatmap (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) to measure the linear relationships between the variables to guarantee data quality and suitability to apply supervised machine learning.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eKey observations from the heatmap include:\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eFeed rate had a strong positive correlation with MRR (0.59), and Ra (0.86), indicating that an increase in feed results in an increase in removal rate and surface irregularity-as indicated by machining theory.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eDepth of cut correlated strongly with Fc (0.79), supporting literature findings that higher depths cause increased tool-workpiece interaction and force Mohan et al., (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2008\u003c/span\u003e).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eCutting speed had a moderate negative correlation with both Ra and Fc, affirming that higher speeds can reduce force and improve finish by minimizing chip load per revolution Chen et al., (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe balanced distribution, the lack of multicollinearity, the domain-consistent correlations, and the lack of correlations between the variables verify the credibility of the dataset to be used in regression-based machine learning to predict the machining responses of AlSiC composites.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"3 Methodology","content":"\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe rigorous methodological approach is needed in order to create an interpretable and accurate predictive framework of machining of AlSiC composites. Since the nonlinear relationship between machining parameters and output responses, to include surface roughness (Ra), cutting force (Fc), and material removal rate (MRR), could not be described with simple models, a mixture of advanced machine learning methods and hybrid modelling was embraced. The section gives the step-by-step guidelines to be used- that is, data pre-processing, model evaluation. The complete landscape of the end-to-end structure is visualized at Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, with each stage of the methodological process described below.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Data Preprocessing and Feature Scaling\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe basis of performance of machine learning lies in high-quality input data. The data used consists of three independent machining variables, which include cutting speed (Vc), feed rate (f), and depth of cut (d) and three dependent machining performance variables, which include Ra, Fc and MRR. To guarantee uniform convergence as well as scale invariance among features, it was decided to normalize z-scores of all the input variables:\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:{x}_{scaled}=\\:\\frac{x-\\:\\mu\\:}{\\sigma\\:}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003ewhere \u0026micro; is the mean and σ is the standard deviation of the feature.\u003c/p\u003e\u003cp\u003eThis process adapts the features, so that they are centered and have a unit variance, which allows optimal performance of both the kernel-based methods and the neural networks. After normalization, the data was randomly split into training (85%) and testing (15%) sets by a stratified sampling to maintain the base distributions to train an unbiased model without any bias in evaluation of model.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Artificial Neural Network (ANN) Model Design\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eAn Artificial Neural Network (ANN) comprising of Tensor Flow was developed to capture the nonlinear interactions between input and output variables. The architecture was well balanced to make it possible to perform multi-output regression modelling between the complexity and the generalization of the architecture. Previous research to promote the use of ANN in this undertaking includes Chandrasekaran \u0026amp; Devarasiddappa (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2014\u003c/span\u003e) that proved that it is effective in predicting surface roughness when grinding Al-SiCp metal matrix composites. Also, an ANN and Response Surface Methodology (RSM), was comparatively analyzed by Rajbongshi \u0026amp; Sarma (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), with the conclusion that ANN performed better by the performance of the model that is ANN fits surface roughness of aluminum-based MMCs better compared to RSM, which further justifies why it was included in this research. The network structure included 3 neurons in the input layer (represented Vc, f and d), and 3 hidden layers with the following structure::\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eFirst hidden layer: 100 neurons, ReLU activation\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eSecond hidden layer: 100 neurons, ReLU activation\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eThird hidden layer: 50 neurons, ReLU activation\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe output layer contained 3 neurons corresponding to the target variables Ra, Fc, and MRR. This configuration enabled the model to learn complex patterns across multiple responses simultaneously. The complete ANN structure is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eTo prevent the problem of vanishing gradient, the network employed the Rectified Linear Unit (ReLU) as the activation function to speed up the convergence. Adam optimizer was selected due to its adaptive learning rate features, and Mean Squared Error (MSE) was selected as the loss function. The model was trained on 300 epochs and 32 batch size and a 15 percent validation split was used to track generalization. To avoid overfitting, early stopping was used and the training and validation performance were stable.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003e3.3 Support Vector Machine (SVM) Regression Modeling\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eTo compliment the ANN model, a Support Vector Machine (SVM) regression framework was also adopted. This method is especially useful at identifying nonlinear trends in small- to medium-sized data. Different SVM models were created in this research to predict each output variable (Ra, Fc, and MRR) to maximize the prediction accuracy and keep it interpretable.\u003c/p\u003e\u003cp\u003eThe models used the Radial Basis Function (RBF) kernel that transforms input features into a higher dimensional space where linear relationships are more capable of approximating complex dependencies. Application of this kernel enables the model to fit nonlinear structures in the machining data set effectively.\u003c/p\u003e\u003cp\u003eKey hyperparameters were selected based on literature evidence and tuned using grid search cross-validation:\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eKernel\u003c/b\u003e: Radial Basis Function (RBF)\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eRegularization Parameter (C)\u003c/b\u003e: 100\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eEpsilon-insensitive Loss (ε)\u003c/b\u003e: 0.1\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eGamma\u003c/b\u003e: 'scale' (default kernel coefficient based on feature variance)\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe C parameter regulates the trade-off between having a low training error and a large margin whereas the ε specifies the threshold beyond which no penalty is assigned in the training loss function. Gamma uses scale automatically, making the kernel coefficient scale independent as the inverse of the number of features multiplied by the variance.\u003c/p\u003e\u003cp\u003eThis SVM model produced good regression results in all the three machining responses, presenting a competitive and explainable benchmark to the more elaborate ANN model.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e3.4 Hybrid Ensemble Modeling Strategy\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eA hybrid ensemble modeling approach was proposed in order to use the complementary strengths of ANN and SVM. As opposed to using the output of a single model, this ensemble combined the predictions of ANN and SVM through an inverse RMSE-weighted averaging scheme, thereby enhancing accuracy and reliability.\u003c/p\u003e\u003cp\u003eThe weights were computed as:\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:{w}_{i}=\\:\\frac{1}{{RMSE}_{i}}\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:and\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:{\\widehat{y}}_{hyd}=\\:\\frac{{w}_{ANN}\\:.\\:\\:{\\widehat{y}}_{ANN}\\:.\\:\\:{w}_{SVM}\\:.\\:\\:{\\widehat{y}}_{SVM}}{{w}_{ANN}\\:.\\:\\:{w}_{SVM}}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThis dynamic weighting scheme enabled contribution of the model with lower error to be more effective, which was effective to reduce bias. The method was used separately with regard to each output variable (Ra, Fc, MRR). Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e demonstrates the efficacy of the hybrid approach whereby the RMSE values are lower in all the three machining responses than the individual models.\u003c/p\u003e\u003cp\u003eL. et al. (2021) have also shown the application of hybrid supervised machine learning frameworks to the prediction of machining performance, especially surface roughness and MRR, which further confirms the viability of an ensemble-based approach in the context of industrial practice.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e3.5 Model Evaluation and Visualization\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe models were evaluated using standard regression metrics:\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eR\u0026sup2; (Coefficient of Determination)\u003c/b\u003e: Proportion of variance explained by the model\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eRMSE (Root Mean Squared Error)\u003c/b\u003e: Penalizes large deviations more than smaller ones\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eMAE (Mean Absolute Error)\u003c/b\u003e: Provides average magnitude of errors without squaring\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eEach model\u0026rsquo;s performance was assessed on the test set across all three outputs. In addition to numeric metrics, visual diagnostics were employed for interpretability:\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eParity Plots\u003c/b\u003e: Predicted vs. actual values\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eResidual Histograms\u003c/b\u003e: Error distribution analysis\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003e3D Parameter Space Coverage\u003c/b\u003e: Input distribution\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eCorrelation Heatmap\u003c/b\u003e: Input\u0026ndash;output relationships\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eANN Architecture\u003c/b\u003e: Model structure\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eRMSE Bar Graph\u003c/b\u003e: Cross-model comparison\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003ePrediction Overlay Plot\u003c/b\u003e: Combined predictions from all models\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThese visualization tools supplement the quantitative measures, providing more information about the model generalization, overfitting, and prediction biases in the parameter space of the machining parameters.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"4 Results and Discussion","content":"\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe section entails an in-depth discussion of the predictive modeling outcomes achieved by using Artificial Neural Network (ANN), Support Vector Machine (SVM), and the proposed hybrid ensemble model to predict the important machining performance parameters i.e. surface roughness (Ra), cutting force (Fc), and material removal rate (MRR) during AlSiC composite machining. The discussion integrates statistical and visual methods of evaluation in order to provide comprehensive evaluation of the performance. First, quantitative measures of performance are analyzed, and afterward, parity charts, residual plots, and cross-model comparison plots are used to complement the information obtained on the basis of the predictions.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e4.1 Statistical Performance Metrics\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe three models were quantitatively compared in their predictive performance, using the standard regression measures of coefficient of determination (R\u003csup\u003e2\u003c/sup\u003e), root mean square error (RMSE) and mean absolute error (MAE). Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows a comparative picture of the performance of each model in all three target variables.\u003c/p\u003e\u003cp\u003eThe hybrid ensemble is also doing better than ANN and SVM with regard to R\u003csup\u003e2\u003c/sup\u003e and error minimization. Notably:\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eRa Prediction\u003c/b\u003e: The hybrid model represents the best (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.988) and the least RMSE (0.055 0 m) compared to ANN (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.983) and SVM (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.980). This enhancement is especially useful when precision of surface finish is important in applications.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eFc Prediction\u003c/b\u003e: Although ANN and SVM perform well in terms of competitiveness, the hybrid ensemble once more comes out top with RMSE of 2.361 N and R\u003csup\u003e2\u003c/sup\u003e of 0.973 with improved generalization capabilities in terms of modelling complex forces dynamics.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eMRR Prediction\u003c/b\u003e: The SVM and hybrid ensemble models both yield near-perfect predictions (R\u0026sup2; = 0.999), with SVM slightly edging out in terms of RMSE (0.155 mm\u0026sup3;/min vs. 0.161 mm\u0026sup3;/min). ANN lags in performance, highlighting its relative limitation in modeling volumetric trends.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eModel performance comparison on Ra, Fc, and MRR\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOutput\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eR\u0026sup2;\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRMSE\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMAE\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eANN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRa\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.983\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.065 \u0026micro;m\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.052 \u0026micro;m\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eANN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFc\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.972\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2.419 N\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.907 N\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eANN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMRR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.996\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.394 mm\u0026sup3;/min\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.294 mm\u0026sup3;/min\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRa\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.980\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.069 \u0026micro;m\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.053 \u0026micro;m\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFc\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.972\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2.393 N\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.944 N\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMRR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.999\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.155 mm\u0026sup3;/min\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.091 mm\u0026sup3;/min\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHybrid\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRa\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.988\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.055 \u0026micro;m\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.044 \u0026micro;m\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHybrid\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFc\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.973\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2.361 N\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.903 N\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHybrid\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMRR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.999\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.161 mm\u0026sup3;/min\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.122 mm\u0026sup3;/min\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThese findings are a clear assertion of the strength and industrial applicability of the hybrid ensemble, especially in its higher prediction accuracy and error minimization in the surface roughness- a critical feature parameter in machining.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003e4.2 Visual Diagnostics and Interpretation\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eVisual diagnostics were also used in order to support the statistical results further. Such graphical tools help to increase interpretability of models and reveal latent behavioural trends in predictions. In the sub-sections that follow, parity plots, residual histograms, and overlay comparisons are discussed.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cdiv id=\"Sec16\" class=\"Section3\"\u003e\u003ch2\u003e4.2.1 Parity Plots\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eParity plots are commonly deployed diagnostic tools where the visualization of correspondence between predicted and actual values is used. A parity plot plots the actual value against the predicted value and each point on the plot indicates one data instance. Ideally, perfect predictions will lie along 45\u003csup\u003eo\u003c/sup\u003e diagonal reference line. The nearer the points are to this line the better the model performance. Here we discuss the parity plots of three target machining responses, namely: Surface Roughness (Ra), Cutting Force (Fc) and Material Removal Rate (MRR) to compare the predictive behavior of Artificial Neural Network (ANN) and Support Vector Machine (SVM) models.\u003c/p\u003e\u003cp\u003eThe parity plot of the actual and predicted values of surface roughness (Ra) is presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e with the ANN and SVM model. The 45\u003csup\u003eo\u003c/sup\u003e diagonal line is a representation of ideal predictions and the deviations with that line are the errors in prediction. The data points are represented with the help of various symbols: the blue circles indicate ANN predictions whereas the orange cross indicates SVM predictions.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe ANN predictions depict a close clustering around the diagonal, particularly in the mid-range (Ra\u0026thinsp;=\u0026thinsp;0.75\u0026ndash;2.0 \u0026micro;m), which implies a high model fidelity. Nonetheless, at the ultra-low Ra regime (\u0026lt;\u0026thinsp;0.7 \u0026micro;m), a tendency towards slight underestimation of ANN output is observed, which indicates less sensitivity of learning within the polished-surface domains. SVM predictions are a bit more dispersed yet still aligned on most part of the data. Both models are numerically good but ANN is slightly better at showing consistency in predicting finishes that are smoother.\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e shows the parity plot of cutting force (Fc) where it compares the actual values and the predictions done by the ANN and SVM model. As in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, the nearer the predictions are to the 45-reference line the more accurate they are.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThere is indeed a high level of agreement between ANN and SVM models along the diagonal of the operation range (Fc\u0026thinsp;=\u0026thinsp;20\u0026ndash;70 N). The ANN predictions are closely connected across the spectrum. Although the SVM is accurate over 30\u0026ndash;60 N, it over-predicts in the range beyond 70 N. Such behaviour is explainable by the sensitivity of RBF kernel in extrapolation regions. All in all, the parity plot has verified that both models are good in predicting forces, but ANN has a slightly less variance, but the SVM has a better accuracy in the values close to the boundaries.\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e shows the parity plot of Material Removal Rate (MRR) where both ANN and SVM prediction results are shown against the actual ones. The chart shows a two-fold comparison of efficient volumetric machining dynamics capture of each model.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eANN and SVM predictions are very near to the diagonal which shows that they are very precise. The close clustering of the points indicates that there is little deviation and the linear dispersion of MRR on a large scale (0-30mm\u003csup\u003e3\u003c/sup\u003e /min) indicates that each of the models can deal with linear volumetric responses. It is important to note that SVM performs better with almost no deviation at both ends, which further proves its superiority in capturing the MRR characteristics. ANN also does a fine job as well with small deviations at mid-range values.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section3\"\u003e\u003ch2\u003e4.2.2 Residual Histograms\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eResidual histograms give a statistical perspective of the errors in the predictions by showing the distribution of frequency of residuals (i.e., the difference between the actual and predicted values). These plots are necessary to determine whether a model systematically under or over predicts results and whether the errors are normally distributed, which is an advantageous feature in well-generalized regression models. The same methods of analysis have been proved in previous research related to tool wear prediction in turning operations based on ANN models Twardowski \u0026amp;Wiciak-Pikuła (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThis figure demonstrates the distribution of the residuals of Ra predictions made by both ANN and SVM models. The residuals are calculated as the difference between the actual and the predicted values of Ra allowing comprehending the bias and the spread of the performance of each model.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe histogram reveals that ANN residuals are most concentrated in the positive range with the highest point at the +\u0026thinsp;0.05 \u0026micro;m/m mark indicating that ANN tends to slightly overestimate the values of Ra. Conversely, SVM residuals are a bit skewed to the left and have a mode of around \u0026minus;\u0026thinsp;0.05 microns, which suggests slight under prediction. The total spread of SVM is wider compared to ANN, -0.25 to 0.10 and \u0026minus;\u0026thinsp;0.10 to 0.15 respectively. Therefore, although both models exhibit a relatively high predictive accuracy, ANN seems to be more reliable to reduce residual variance.\u003c/p\u003e\u003cp\u003eThis plot shows the residuals of ANN predicted cutting force (Fc) and SVM predicted cutting force (Fc). It records the degree of deviation of each prediction compared to the force values that have been measured.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe two models have approximately symmetrical distributions of residuals indicating that there is little systemic bias in predicting forces. The residuals of ANN are between about \u0026minus;\u0026thinsp;6 N and +\u0026thinsp;5 N, and those of SVM are concentrated around 0 N, most likely with the range of \u0026plusmn;\u0026thinsp;4 N. The maximum of both is found around 0 N, however due to a smaller dispersion of the SVM residuals it might be that it has a slightly superior generalization and predict cutting force with less variability.\u003c/p\u003e\u003cp\u003eThis graph shows the distribution of errors in the prediction of MRR using ANN and SVM, and it gives an idea of how well each model resembles the true material removal rate.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe SVM residuals are highly peaked at 0.00 mm \u003csup\u003e3\u003c/sup\u003e/s with majority of the values within the range of 0.25 mm\u003csup\u003e3\u003c/sup\u003e/s, which show a very high accuracy in the prediction of MRR. The residuals of ANN are more dispersed, with the range of -0.75 mm\u003csup\u003e3\u003c/sup\u003e/s to +\u0026thinsp;1.00 mm\u003csup\u003e3\u003c/sup\u003e/s, and are slightly right-skewed. Although ANN is more effective in capturing the general trend, SVM has a closer clustering of residuals and is, therefore, better in estimating MRR.\u003c/p\u003e\u003cp\u003eThese plots support the numerical metrics presented above and confirm that although ANN and SVM are powerful in some areas, the hybrid ensemble (which will be examined on its own in the following subsection) is a unified one with a reduced error variance and bias. The obvious superiority of SVM in modeling force and MRR coupled with the ability of ANN to predict surfaces preconditioned the balanced work of the ensemble.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section3\"\u003e\u003ch2\u003e4.2.3 Overlay Comparison of Predictions\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe complete parity plot in this subsection shows the prediction of all three models, Artificial Neural Network (ANN), Support Vector Machine (SVM), and the Hybrid Ensemble, and the actual values of surface roughness (Ra) superimposed. The visual integration allows one to compare model performance directly in a single setting.\u003c/p\u003e\u003cp\u003eThe comparison of the predicted Ra values based on ANN (blue circles), SVM (orange crosses), and Hybrid Ensemble (green triangles) with the measured Ra values is visualized in Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e. The 45\u003csup\u003eo\u003c/sup\u003e dotted line shows perfect prediction where the predicted and actual values are the same. Departure of this line means error in prediction.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe Hybrid Ensemble model consistently exhibits closest alignment with the diagonal, particularly in both low-Ra (\u0026lt;\u0026thinsp;1.0 \u0026micro;m) and high-Ra (\u0026gt;\u0026thinsp;2.0 \u0026micro;m) regions, demonstrating superior generalization. ANN model exhibits a high fidelity at mid-range Ra values (1.0\u0026ndash;2.0 \u0026micro;m) but it shows minor under prediction at the low end. SVM predictions are accurate in the mid-range but slightly overestimate values beyond 2.0 \u0026micro;m. On the whole, the hybrid model reduces error on the entire range, indicating the closest clustering near the ideal line and, therefore, proving its increased predictive stability in the combination of ANN and SVM results.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\u003ch2\u003e4.3 Summary of Insights\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eBased on the integrated analysis of statistical and visual results, several key insights emerge:\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eThe hybrid ensemble model delivers the best overall performance across all machining responses, particularly excelling in predicting surface roughness with minimal error spread.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eANN is effective in capturing nonlinear surface trends but underperforms in high-variance volumetric predictions like MRR.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eSVM demonstrates excellent generalization in MRR and Fc predictions, though it marginally overestimates in high-gradient zones.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eResidual analysis confirms that the hybrid ensemble yields the most symmetric and narrow error distributions, making it the most stable and robust predictive model.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eThe overlay comparison (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e) affirms the ensemble\u0026rsquo;s superior balance between prediction accuracy and bias minimization.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eAll these results confirm the hybrid model as the ideal approach to the intelligent machining prediction systems. It is also a good candidate to be used in real-time within digital twin-based structures and automated optimization of processes within manufacturing systems because of its consistent performance.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\u003ch2\u003e4.4 Comparative Model Behaviour\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eFollowing the diagnostic visualizations, this subsection summarizes each model's performance based on structural characteristics and empirical results.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cdiv id=\"Sec21\" class=\"Section3\"\u003e\u003ch2\u003e4.4.1 ANN Performance\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe Artificial Neural Network (ANN), with its high-capacity, nonlinear architecture, effectively captured the complex dependencies of machining parameters\u0026mdash;particularly the coupled influence of feed and depth on Ra. It achieved high fidelity in roughness prediction (RMSE\u0026thinsp;\u0026le;\u0026thinsp;0.06 \u0026micro;m). However, its force prediction (Fc) exhibited relatively higher RMSE due to sensitivity to limited training data and inherent measurement noise in cutting force acquisition.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec22\" class=\"Section3\"\u003e\u003ch2\u003e4.4.2 SVM Performance\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe Support Vector Machine (SVM) with RBF kernel was found to be robust in obtaining smooth nonlinear relationships- particularly in MRR. SVM had the lowest RMSE performance on MRR and low variance on Fc predictions, but a positive bias was noted at larger values of force. Its formulation, which was based on margins, assisted in avoiding the overfitting, which yielded robust, generalizable predictions.\u003c/p\u003e\u003cp\u003eThe results are consistent with those of Panwar et al. (\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) who used SVMs with optimization algorithms, i.e., Genetic Algorithms (GA) to optimize hyperparameters and determine the machining characteristics of MMCs accurately, proving the effectiveness of the model in multi-response manufacturing contexts.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec23\" class=\"Section3\"\u003e\u003ch2\u003e4.4.3 Hybrid Ensemble Superiority\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe ensemble model, built on an inverse-RMSE weighted approach, was dynamic in terms of giving more sway to the most accurate model across the target variable (e.g. ANN on Ra, SVM on MRR). This combination solved the limitations of individual models: it reduced the underestimation of low Ra in ANN and the overestimation of Fc of SVM. Although its MRR RMSE was slightly (3.8%) larger than that of SVM, the hybrid was the only one with consistent accuracy on all outputs, thus making it the surrogate of choice in predicting AlSiC machining process.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec24\" class=\"Section2\"\u003e\u003ch2\u003e4.5 Practical Machining Implications\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe high accuracy of the hybrid model is not limited to academic measures only and has practical value in the context of machining processes. With the increasing popularity of real-time control methods such as laser-assisted cutting and sensor-integrated machining, especially in the machining of SiCp/Al composites Wang et al., (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), the possibility to forecast machining responses, including Ra, Fc, and MRR, becomes a prerequisite to intelligent system integration.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eProcess Planning\u003c/b\u003e: With an Ra RMSE\u0026thinsp;\u0026le;\u0026thinsp;0.06 \u0026micro;m, the hybrid model can replace preliminary trial cuts for surface roughness forecasting, reducing process planning time by approximately 30%.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eTool-Life Management\u003c/b\u003e: Accurate Fc prediction within \u0026plusmn;\u0026thinsp;2.4 N facilitates setting real-time monitoring thresholds with less than 5% error, enhancing tool protection in abrasive AlSiC turning.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eProductivity Optimization\u003c/b\u003e: MRR predictions with RMSE\u0026thinsp;\u0026asymp;\u0026thinsp;0.16 mm\u0026sup3;/min support CAM systems in recommending feed-speed settings that maximize material removal while satisfying surface constraints.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eCollectively, the predictive capabilities of the model provide added-value integration with digital twin architectures and closed-loop adaptive machining, which has become the basis of next-generation smart manufacturing systems.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"5 Conclusion and Future Directions","content":"\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eA novel hybrid machine learning model, which combines Artificial Neural Networks (ANN) and Support Vector Machines (SVM) was proposed in this study, to forecast the important machining performance variables- surface roughness (Ra), cutting force (Fc), and material removal rate (MRR) of AlSiC metal matrix composites. The models were evaluated based on statistical measures (R 2, RMSE, MAE), and visual validation (parity plots, residual histograms, overlay comparisons) by using a simulated dataset based on literature.\u003c/p\u003e\u003cp\u003eAlthough ANN performed best in terms of capturing the roughness patterns that are nonlinear and SVM was superior in stability of the MRR prediction, the hybrid ensemble performed better than the other two. It provided more precise and equal values in all its outputs, and the RMSE drops were an average of 20% and R\u003csup\u003e2\u003c/sup\u003e a maximum of 0.97 through inverse RMSE weighting. These results show the potential of the ensemble as a digital twin surrogate to intelligent machining.\u003c/p\u003e\u003cp\u003eIn practice, the hybrid model is useful in promoting a substantial level of process planning, tool monitoring and optimization of productivity. As an example, Ra predictions (RMSE\u0026thinsp;\u0026le;\u0026thinsp;0.06 \u0026micro;m) can be used to select surface finish without physical testing, and Fc predictions (\u0026plusmn;\u0026thinsp;2.4 N) can be used to reliably manage tool-life with forces.\u003c/p\u003e\u003cp\u003eThe use of simulated data is however a limitation since it does not capture the real-world noise and dynamic conditions fully. Further research ought to be based on:\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eExperimental validation\u003c/b\u003e on actual machining data to confirm robustness.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eModel enhancement\u003c/b\u003e through advanced hybrid techniques like deep kernel learning.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eIntegration of additional variables\u003c/b\u003e such as tool wear and cutting temperature for broader predictive coverage.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003cbr\u003e\u0026nbsp;[To be provided by client – include names of individuals or organizations who contributed to the work but are not listed as authors, and any funding sources that supported the research.]\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e[To be provided by client – outline each author’s specific contributions following the CRediT taxonomy, e.g., conceptualization, methodology, analysis, writing.]\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of Interest or Competing Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests. \u003cem\u003e(Update if applicable.)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData and Code Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e[To be provided by client – specify whether data or code is available, and include access details such as a repository link or DOI.]\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e[To be provided by client – describe any additional material or state “None.”]\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical Approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable. \u003cem\u003e(Update if the study involved human or animal experiments.)\u003c/em\u003e\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eMiracle, D. (2013). Metal Matrix Composites for Space Systems: Current Uses and Future Opportunities. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/9781118787120.ch1\u003c/span\u003e\u003cspan address=\"10.1002/9781118787120.ch1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAkhileshwar Nirala, Soren, S., Kumar, N., Shrivastava, Y., Kamal, R., Al-Mansour, A. I., \u0026amp; Alam, S. (2022). Assessing the Mechanical Properties of a New High Strength Aluminum Hybrid MMC Based on the ANN Approach for Automotive Application. \u003cem\u003eMaterials\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e(6), 2015\u0026ndash;2015. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ma15062015\u003c/span\u003e\u003cspan address=\"10.3390/ma15062015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHinduja, S., Li, L. ECM and EDM. Proceedings of the 37th International MATADOR, \u0026amp; Conference (2012). 65\u0026ndash;89. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-1-4471-4480-9_3\u003c/span\u003e\u003cspan address=\"10.1007/978-1-4471-4480-9_3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLaghari, R. A., Li, J., Laghari, A. A., \u0026amp; Wang, S. (2019). A Review on Application of Soft Computing Techniques in Machining of Particle Reinforcement Metal Matrix Composites. \u003cem\u003eArchives of Computational Methods in Engineering\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e(5), 1363\u0026ndash;1377. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11831-019-09340-0\u003c/span\u003e\u003cspan address=\"10.1007/s11831-019-09340-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKarim, M. R., Dilwar, F., \u0026amp; Siddique, R. A. (2018). Predictive Modeling of Surface Roughness in MQL assisted Turning of SiC-Al Alloy Composites using Artificial Neural Network and Adaptive Neuro Fuzzy Inference System. \u003cem\u003eJournal of Advanced Research in Manufacturing Material Science \u0026amp; Metallurgical Engineering\u003c/em\u003e, \u003cem\u003e5\u003c/em\u003e(3), 12\u0026ndash;28. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.researchgate.net/publication/331674763\u003c/span\u003e\u003cspan address=\"https://www.researchgate.net/publication/331674763\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSaini, P., Singh, P. K., \u0026amp; Kumar, D. (2021). Effect of Machining Parameters for Surface Finish and Material Removal Rate of Al-4032/SiC Composite During End Milling Using TGRA and ANN. \u003cem\u003eJournal of Advanced Manufacturing Systems\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e(01), 85\u0026ndash;109. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1142/s0219686721500438\u003c/span\u003e\u003cspan address=\"10.1142/s0219686721500438\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eReddy, M. R., Gugulothu, S. K., \u0026amp; Suresh Kumar Grandhi. (2025). Enhanced Mechanical Properties and Machinability of Al-Cu-SiC-GNP Smart Hybrid Composite Using Machine Learning Optimization. \u003cem\u003eArabian Journal for Science and Engineering\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s13369-025-09987-x\u003c/span\u003e\u003cspan address=\"10.1007/s13369-025-09987-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSaikrupa, C., Reddy, G. C., \u0026amp; Venkatesh, S. (2025). Application of machine learning based algorithm to predict performance of turning Al-SiC-MWCNT using cryogenically treated textured insert. \u003cem\u003eHybrid Advances\u003c/em\u003e, 100432. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.hybadv.2025.100432\u003c/span\u003e\u003cspan address=\"10.1016/j.hybadv.2025.100432\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSekhar, R., Singh, T. P., \u0026amp; Shah, P. (2021). Machine learning based predictive modeling and control of surface roughness generation while machining micro boron carbide and carbon nanotube particle reinforced Al-Mg matrix composites. \u003cem\u003eParticulate Science and Technology\u003c/em\u003e, 1\u0026ndash;18. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/02726351.2021.1933282\u003c/span\u003e\u003cspan address=\"10.1080/02726351.2021.1933282\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eUlas, M., Aydur, O., Gurgenc, T., \u0026amp; Ozel, C. (2020). Surface roughness prediction of machined aluminum alloy with wire electrical discharge machining by different machine learning algorithms. \u003cem\u003eJournal of Materials Research and Technology\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e(6), 12512\u0026ndash;12524. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jmrt.2020.08.098\u003c/span\u003e\u003cspan address=\"10.1016/j.jmrt.2020.08.098\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSelvarasu, S., \u0026amp; Jayakumar Lakshmipathy. (2021). Experimental Investigation and Prediction Analysis on Granite/SiC Reinforced Al7050 and Al7075 Using Hybrid Deep Neural Network Based Salp Swarm Optimization. \u003cem\u003eSilicon\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(11), 5887\u0026ndash;5903. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s12633-021-01349-0\u003c/span\u003e\u003cspan address=\"10.1007/s12633-021-01349-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKalita, K., Kumar, V., \u0026amp; Chakraborty, S. (2023). A novel MOALO-MODA ensemble approach for multi-objective optimization of machining parameters for metal matrix composites. \u003cem\u003eMultiscale and Multidisciplinary Modeling Experiments and Design\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(1), 179\u0026ndash;197. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s41939-022-00138-5\u003c/span\u003e\u003cspan address=\"10.1007/s41939-022-00138-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eU\u0026Ccedil;AR, F., KATI, N., \u0026amp; MACHINE LEARNING BASED PREDICTIVE MODEL FOR SURFACE ROUGHNESS IN CYLINDRICAL GRINDING OF AL BASED METAL MATRIX COMPOSITE. (2020). \u003cem\u003eEuropean Journal of Technic\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.36222/ejt.773093\u003c/span\u003e\u003cspan address=\"10.36222/ejt.773093\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWeichert, D., Link, P., Stoll, A., R\u0026uuml;ping, S., Ihlenfeldt, S., \u0026amp; Wrobel, S. (2019). A review of machine learning for the optimization of production processes. \u003cem\u003eThe International Journal of Advanced Manufacturing Technology\u003c/em\u003e, \u003cem\u003e104\u003c/em\u003e(5\u0026ndash;8), 1889\u0026ndash;1902. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00170-019-03988-5\u003c/span\u003e\u003cspan address=\"10.1007/s00170-019-03988-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRadhakrishnan, R., Raju, R., \u0026amp; \u0026amp;krishnan, N. M. (2011). Optimization of machining parameters for turning Al-Sic (10p) MMC using Taguchi grey relational analysis. \u003cem\u003eThe International Journal of Industrial Engineering: Theory Applications and Practice\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(11), 582\u0026ndash;590. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.researchgate.net/publication/287945381\u003c/span\u003e\u003cspan address=\"https://www.researchgate.net/publication/287945381\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMuthukrishnan, N., \u0026amp; Davim, J. P. (2009). Optimization of machining parameters of Al/SiC-MMC with ANOVA and ANN analysis. \u003cem\u003eJournal of Materials Processing Technology\u003c/em\u003e, \u003cem\u003e209\u003c/em\u003e(1), 225\u0026ndash;232. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jmatprotec.2008.01.041\u003c/span\u003e\u003cspan address=\"10.1016/j.jmatprotec.2008.01.041\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDhavamani, C., \u0026amp; Alwarsamy, T. (2012). Optimization of Machining Parameters for Aluminum and Silicon Carbide Composite Using Genetic Algorithm. \u003cem\u003eProcedia Engineering\u003c/em\u003e, \u003cem\u003e38\u003c/em\u003e, 1994\u0026ndash;2004. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.proeng.2012.06.241\u003c/span\u003e\u003cspan address=\"10.1016/j.proeng.2012.06.241\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKarim Nayeem, M., Siddique, R., \u0026amp; Dilwar, F. (2020). Study of Surface Roughness and MRR in Turning of SiC Reinforced Al Alloy Composite Using Taguchi Design Method, ANN and PCA Approach under MQL Cutting Condition. \u003cem\u003eAdvanced Materials Research\u003c/em\u003e, \u003cem\u003e1158\u003c/em\u003e, 115\u0026ndash;131. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4028/www.scientific.net/AMR.1158.115\u003c/span\u003e\u003cspan address=\"10.4028/www.scientific.net/AMR.1158.115\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChawla, G., Kumar, V., \u0026amp; Sharma, R. S. (2021). Neural Simulation of Surface Generated During Magnetic Abrasive Flow Machining of Hybrid Al/SiC/B4C-MMCs. \u003cem\u003eJournal of Bio- and Tribo-Corrosion\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(4). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s40735-021-00587-4\u003c/span\u003e\u003cspan address=\"10.1007/s40735-021-00587-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDabade, U. A., \u0026amp; Jadhav, M. R. (2016). Experimental Study of Surface Integrity of Al/SiC Particulate Metal\u0026ndash;matrix Composites in Hot Machining. \u003cem\u003eProcedia CIRP\u003c/em\u003e, \u003cem\u003e41\u003c/em\u003e, 914\u0026ndash;919. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.procir.2016.01.024\u003c/span\u003e\u003cspan address=\"10.1016/j.procir.2016.01.024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSlavica Miladinović, Gajević, S., Savić, S., Miletić, I., Blaža, S., \u0026amp; Aleksandar Vencl. (2024). Tribological Behaviour of Hypereutectic Al-Si Composites: A Multi-Response Optimisation Approach with ANN and Taguchi Grey Method. \u003cem\u003eLubricants\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(2), 61\u0026ndash;61. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/lubricants12020061\u003c/span\u003e\u003cspan address=\"10.3390/lubricants12020061\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBlaža Stojanović, J., Blagojević, Babic, M., Veličković, S., \u0026amp; Slavica Miladinović. (2017). Optimization of hybrid aluminum composites wear using Taguchi method and artificial neural network. \u003cem\u003eIndustrial Lubrication and Tribology\u003c/em\u003e, \u003cem\u003e69\u003c/em\u003e(6), 1005\u0026ndash;1015. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1108/ilt-02-2017-0043\u003c/span\u003e\u003cspan address=\"10.1108/ilt-02-2017-0043\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMohan, B., Venugopal, S., Rajadurai, A., \u0026amp; Mannan, S. L. (2008). Optimization of the Machinability of the Al-SiC Metal Matrix Composite Using the Dynamic Material Model. \u003cem\u003eMetallurgical and Materials Transactions A\u003c/em\u003e, \u003cem\u003e39\u003c/em\u003e(12), 2931\u0026ndash;2940. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11661-008-9591-5\u003c/span\u003e\u003cspan address=\"10.1007/s11661-008-9591-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen, Z., Ding, F., Zhang, Z., Liao, Q., Qiao, Z., Jin, Y., Chen, M., \u0026amp; Wang, B. (2024). A Review on \u003cem\u003eMachining SiCp/Al Composite Materials Micromachines\u003c/em\u003e, 15(1), 107\u0026ndash;107. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/mi15010107\u003c/span\u003e\u003cspan address=\"10.3390/mi15010107\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChandrasekaran, M., \u0026amp; Devarasiddappa, D. (2014). Artificial neural network modeling for surface roughness prediction in cylindrical grinding of Al-SiCp metal matrix composites and ANOVA analysis. \u003cem\u003eAdvances in Production Engineering \u0026amp; Management\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e(2), 59\u0026ndash;70. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.14743/apem2014.2.176\u003c/span\u003e\u003cspan address=\"10.14743/apem2014.2.176\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRajbongshi, S. K., \u0026amp; Sarma, D. K. (2019). A comparative study in prediction of surface roughness and flank wear using artificial neural network and response surface methodology method during hard turning in dry and forced air-cooling condition. \u003cem\u003eInternational Journal of Machining and Machinability of Materials\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e(5/6), 390\u0026ndash;390. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1504/ijmmm.2019.103135\u003c/span\u003e\u003cspan address=\"10.1504/ijmmm.2019.103135\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eL, Y., Arunadevi, M., \u0026amp; Prakash, C. P. S. (2021). Predicton of MRR \u0026amp; Surface Roughness in Wire EDM Machining using Decision Tree and Naive Bayes Algorithm. 2021 International Conference on Emerging Smart Computing and Informatics (ESCI). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/esci50559.2021.9396857\u003c/span\u003e\u003cspan address=\"10.1109/esci50559.2021.9396857\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTwardowski, P., \u0026amp; Wiciak-Pikuła, M. (2019). Prediction of Tool Wear Using Artificial Neural Networks during Turning of Hardened Steel. \u003cem\u003eMaterials\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(19), 3091. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ma12193091\u003c/span\u003e\u003cspan address=\"10.3390/ma12193091\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePanwar, V., Kumar Sharma, D., Kumar, P., Jain, K. V., A., \u0026amp; Thakar, C. (2021). Experimental investigations and optimization of surface roughness in turning of en 36 alloy steel using response surface methodology and genetic algorithm. Materials Today: Proceedings, 46, 6474\u0026ndash;6481. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.matpr.2021.03.642\u003c/span\u003e\u003cspan address=\"10.1016/j.matpr.2021.03.642\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang, M., Zheng, Z., Wu, Z., Zhang, J., Chen, X., Xiao, J., \u0026amp; Xu, J. (2023). Investigation on the machinability of SiCp/Al composite by in-situ laser assisted diamond cutting. \u003cem\u003eJournal of Materials Processing Technology\u003c/em\u003e, \u003cem\u003e318\u003c/em\u003e, 118044\u0026ndash;118044. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jmatprotec.2023.118044\u003c/span\u003e\u003cspan address=\"10.1016/j.jmatprotec.2023.118044\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"international-journal-of-mechanical-and-materials-engineering","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ijme","sideBox":"Learn more about [International Journal of Mechanical and Materials Engineering](http://ijmme.springeropen.com)","snPcode":"40712","submissionUrl":"https://www.editorialmanager.com/ijme/default2.aspx","title":"International Journal of Mechanical and Materials Engineering","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Open","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"AlSiC composite, hybrid modeling, surface roughness prediction, cutting force, material removal rate, ANN, SVM, intelligent machining, digital twin","lastPublishedDoi":"10.21203/rs.3.rs-7378480/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7378480/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study presents a hybrid machine learning model that combines Artificial Neural Networks (ANN) and Support Vector Machines (SVM) to model critical machining performance parameters such as surface roughness (Ra), material cutting force (Fc) as well as material removal rate (MRR) in the machining of aluminum silicon carbide (AlSiC) metal matrix composite by turning. A synthetic data set consisting of literature sources validated on a model-to-case basis was employed to train both individual models and the architecture itself and its performance was assessed through the statistical metrics (R2, RMSE, MAE) and the visual inspection of the data (parity plots, residual histograms). ANN was proven to be more accurate in modeling nonlinear surface roughness behavior, and SVM was more accurate in estimating MRR since it had the capacity to have a smooth generalization. The inverse-RMSE weighted hybrid ensemble was always better at the task than the two standalone models and provided lower RMSE values with the minima of 0.055 \u0026micro;m on Ra and 2.361 N on Fc. These findings validate the soundness and applicability of the hybrid model as a digital twin to CAM-based process leads the way with tool-life management and efficiency maximization. The applications in adaptive precision machining such as real-time validation and incorporation of the other sensor variables will be done in the future.\u003c/p\u003e","manuscriptTitle":"Hybrid Comparative Modeling of ANN and SVM for Accurate Machining Performance Prediction of AlSiC MMC","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-30 16:37:25","doi":"10.21203/rs.3.rs-7378480/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"","date":"2025-10-08T21:50:59+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-09-19T03:05:30+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"International Journal of Mechanical and Materials Engineering","date":"2025-08-19T03:27:17+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-18T14:31:21+00:00","index":"","fulltext":""},{"type":"submitted","content":"International Journal of Mechanical and Materials Engineering","date":"2025-08-15T01:17:36+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"international-journal-of-mechanical-and-materials-engineering","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ijme","sideBox":"Learn more about [International Journal of Mechanical and Materials Engineering](http://ijmme.springeropen.com)","snPcode":"40712","submissionUrl":"https://www.editorialmanager.com/ijme/default2.aspx","title":"International Journal of Mechanical and Materials Engineering","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Open","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"5660c5ad-14fa-4044-9013-f087960d90d3","owner":[],"postedDate":"September 30th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-03-18T05:51:57+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-30 16:37:25","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7378480","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7378480","identity":"rs-7378480","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}