Machine learning- driven computational analysis for predicting electrical properties of polypyrrole composites

Developing sustainable conducting polymers through green chemistry offers a promising route toward environmentally responsible electronic materials. In this study, polypyrrole (PPy) composites were synthesized using an eco-friendly approach that employed lemon (Citrus limon) extract as a natural dopant and latex from Calotropis gigantea (Arka) as a bio-stabilizer. The effect of varying dopant volumes (0.5–3 mL) on the frequency-dependent electrical response of the composites was systematically examined across the range of 100 Hz to 1 MHz. Experimental results showed pronounced variations in electrical conductivity, dielectric constant, and dielectric loss with changes in dopant concentration and frequency. To complement these findings, machine learning (ML) models—CatBoost, Random Forest, and XGBoost—were trained using compositional and frequency data to predict the electrical properties of both tested and untested systems. The trained models were further applied to predict the electrical properties of untested compositions, including PPy with untested volumes thereby extending the analysis beyond the experimentally measured systems. Five-fold cross-validation confirmed model reliability, while SHapley Additive exPlanations (SHAP) analysis provided insights into the relative influence of each input variable. Among the tested algorithms, CatBoost achieved the highest predictive accuracy (R² > 0.95) with minimal errors across all three target parameters, closely aligning with experimental observations. This integrated approach highlights the synergy between green synthesis, predictive modeling, and explainable AI in designing PPy-based materials with tunable dielectric and electrical properties for sustainable electronic applications. • Machine learning models were used to train electrical properties of polypyrrole composites. • CatBoost, XGBoost and Random Forest were applied for training and testing of electrical property data. • Trained models were used to predict electrical properties for new composite volume fractions. • Models can be expanded with data from other conducting polymers like PTh, PANI and polyethylene. • Enables prediction of co-polymer composite properties without extensive experimental work.