Fig. 1: Workflow of the ML-accelerated HER catalytic activity prediction.
The process includes data collection, feature extraction, model training, model building, feature analysis, and model prediction. Data collection: collect the atomic structures and ∆GH for HER. Feature extraction: extract structural features, electronic features, and atomic features from the atomic structures of the HER. Model training: improve ML models accuracy through hyperparameter tuning. Model building: use the ML models, fitted on the training set, to make predictions on the test set. Feature analysis: analyze feature importance and correlations, and use feature engineering to reduce the feature set while introducing key features to enhance model accuracy. Model prediction: use the ML model to predict potential HECs.
