Predictive modelling of hydrogen adsorption energy using hybrid EMT and machine learning techniques

This study evaluates the predictive performance of standalone and hybrid models for estimating hydrogen adsorption energy, a key parameter in catalysis and surface science. Specifically, the Effective Medium Theory (EMT), Random Forest (RF), and XGBoost (XGB) models were assessed individually and in hybrid configurations using both simulated and real datasets. Model performance was evaluated using standard accuracy metrics, including Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R2). The simulation results indicate that both EMT and XGBoost models perform competitively, with XGBoost achieving the highest R2 value of 0.9432. Hybrid models demonstrated improved predictive performance in simulated data, while the ensemble machine learning model (RF + XGBoost) achieved superior accuracy on real data. The results further show that EMT enhances predictive capability when integrated with machine learning models, whereas purely data-driven ensemble approaches exhibit stronger generalization performance on real datasets. The overall findings highlight the effectiveness of combining physics-based and data-driven approaches for improving predictive accuracy and reliability in adsorption energy estimation. The study recommends the integration of hybrid modelling frameworks into high-throughput materials screening workflows and suggests further investigation into model interpretability, uncertainty quantification, and generalizability.