Developing a novel hybrid ensemble machine learning model for predicting reduced elongation of low-carbon steel in hydrogen environments

Hydrogen embrittlement (HE) poses a substantial challenge to the structural integrity and ductility of low-carbon steel, particularly in the context of hydrogen transport infrastructure. This study introduces a pioneering hybrid ensemble machine learning (ML) model specifically developed to predict the reduced elongation of low-carbon steel in a hydrogen environment with enhanced accuracy and robustness. Seven individual ML models were developed, including Random Forest, Decision Tree, K-Nearest Neighbors (KNN), Gradient Boosting, CatBoost, XGBoost, and Support Vector Regression (SVR). A novel model selection framework that holistically integrates feature importance, R 2 , RMSE, and MAE metrics was used to identify Gradient Boosting and KNN as the top-performing models. These were subsequently combined using a weighted soft voting approach to develop a new hybrid ensemble model. The resulting hybrid model achieved a training R 2 of 0.89 and a test R 2 of 0.87, with RMSE and MAE values of 4.45 and 3.16, respectively. This research represents the first comprehensive integration of hybrid ensemble learning for predicting hydrogen-induced ductility loss in steels, providing a powerful data-driven solution to enhance the reliability of hydrogen transport infrastructure.