Modeling Dynamic Viscosity of Pure Fatty Acid Methyl Esters via Robust Machine Learning Approaches

Fatty acid methyl esters (FAMEs) are renewable, biodegradable biofuels. This study built machine learning models (DT, AdaBoost, EL, K ‐nearest neighboring ( K NN), random forest (RF), EN, CNN, SVR, and MLP‐ANN) to predict their dynamic viscosity using 488 literature data points. Inputs: temperature, pressure, molar mass, and C/H/O fractions. Models used five‐fold cross‐validation (90% training/validation, 10% testing). Metrics: R 2 , mean squared error (MSE), AARE%. MLP‐ANN performed best ( R 2 = 0.9987, MSE = 0.0038, and AARE = 1.68%). DT and AdaBoost showed higher errors; elastic net was weakest ( R 2 = 0.8486, AARE ≈ 25.78%). Pressure was the most impactful parameter, followed by temperature. SHapley Additive exPlanations (SHAP) analysis confirmed pressure as dominant. The framework is robust, accurate, and cost‐effective across broad thermodynamic conditions.