Prediction of Thermodynamic Properties of Sacred Pepper (Piper auritum) Using Machine Learning

This study compares the application of three Machine Learning (ML) models—Artificial Neural Networks (ANNs), Random Forest (RF), and Support Vector Machines (SVM)—for predicting thermodynamic properties (enthalpy change, ΔH; entropy change, ΔS; and Gibbs free energy change, ΔG) of Piper auritum . The models were developed using experimental data correlating these properties with moisture content and temperature. Each model’s architecture was optimized: ANNs with a three-layer feedforward structure trained with the Levenberg-Marquardt algorithm, RF with property-specific tree counts, and SVM with a Radial Basis Function kernel. Model performance was evaluated using k-fold cross-validation and metrics including R², MAE, MSE, and RMSE. Results show that all three ML approaches effectively capture nonlinear relationships, but ANNs outperformed the others, achieving high R² values (e.g., overall R² > 0.999 for ΔH and ΔS). SVM provided intermediate performance, while RF was the least accurate. ANNs are the most powerful and reliable tool for this regression task, offering a highly accurate computational method to predict thermodynamic behavior and potentially reduce experimental characterization needs.