Predictive modeling of thermoplastic nanocomposites using machine learning algorithms
Harshit Sharma, Gaurav Arora, Papiya Bhowmik, Manoj Kumar Singh, Vinod Ayyappan, Anuj Kumar Sehgal, Sanjay Mavinkere Rangappa, Suchart Siengchin
Abstract
In this investigation, various machine learning (ML) algorithms, i.e. Support Vector Regression (SVR), Decision Tree (DT), Random Forest (RF), and k-Nearest Neighbors (kNN) were employed to estimate the mechanical properties of microwave-processed polymer nanocomposites. A universal testing machine evaluated the tensile strength, fracture toughness, and Young’s modulus of Polypropylene/Carbon Nanotubes (PPNT), High-Density Polyethylene/Carbon Nanotubes (HPNT), and Low-Density Polyethylene/Carbon Nanotubes (LPNT) composites. The results indicate that SVR is the most accurate and reliable model for estimating the mechanical properties of all composites. Its superior performance is due to its ability to minimize inaccuracies and generalize effectively, even with limited datasets. Among all models, SVR consistently delivered the best performance for all three mechanical characteristics. It achieved low values for root mean square error (RMSE), mean absolute error (MAE), and mean square error (MSE) i.e. 0.0228, 0.0189, and 0.0005, respectively along with a high coefficient of determination (R 2 ) of 0.9875. These metrics highlight its low error, excellent generalization, low bias, and consistent performance across all datasets. Despite RF performing better than DT, it underperformed compared to SVR, with RMSE variations of 6.1% for HPNT, 47.3% for PPNT, and 32.4% for LPNT. Meanwhile, kNN proved to be the least suitable algorithm in this study due to its poor predictive accuracy and lack of stability.