Comparative analysis of machine learning models for predicting dielectric properties in MoS2 nanofiller-reinforced epoxy composites

Abstract

This research investigates the dielectric properties of nano epoxy composites by incorporating various concentrations of MoS2 into epoxy resin. The study explores the impact of synthesized nanoparticles on undoped epoxy composites, specifically focusing on their potential applications in dielectric materials. The experimental synthesis and characterization of nanoepoxy composites typically involve time-consuming and expensive methods. This study compares five machine learning (ML) models—random forests, decision trees, extra trees, XGBoost, and gradient boosting—in order to predict the frequency-dependent dielectric constants in these composites under different nanofiller variations in order to address this challenge. To ensure robust model performance, training is carried out on different subsets of the dataset, ranging from 60% to 30%, while the remaining portions are reserved for testing purposes (40% to 70%). The main objective of the study is to assess the performance of each regressor technique using metrics such as adjusted R2 score, MSE, RMSE, and MAE, in which the ET regressor excels. The ET method demonstrates exceptional performance, achieving an adjusted R2 value of 0.9977 and 0.9912 for target variables ε′ and ε′′, respectively when tested with a size of 0.4. The findings underscore the potential of ML models for precise and efficient prediction of frequency-dependent dielectric constants of nanoepoxy composites with various concentrations of nanofillers, offering an alternative to time-consuming and expensive laboratory work.