An Interpretable and Streamlined Machine Learning Framework for Simultaneous Prediction of Polarization Curves and Maximum Power Density in Direct Methanol Fuel Cells

A streamlined machine learning framework is developed to predict the polarization curve and maximum power density of direct methanol fuel cells (DMFCs) utilizing Nafion–mordenite composite membranes. A single data set comprising 135 experimental polarization curves, spanning a wide range of practical operating conditions, is used to predict both polarization behavior and power output without requiring separate data sets. Each polarization curve is segmented into activation, Ohmic, and mass-transport loss regions using derivative-based analysis, enabling physically interpretable modeling through well-defined transition parameters. Four practical input variables are considered, including proton conductivity, methanol permeability, methanol concentration, and temperature. Extreme gradient boosting (XGB) regression demonstrates superior predictive performance compared with multiple linear regression, indicating nonlinear interactions governing DMFC performance. The proposed framework achieves high accuracy for polarization curve prediction (R2 = 0.963) and maximum power density prediction (R2 = 0.849). Sensitivity analysis confirmed that membranes optimized for high proton conductivity and low methanol permeability consistently maximize the power density across diverse temperatures. This framework provides a precise and physically interpretable tool for optimizing the membrane design and operating conditions.