Hybrid machine learning with optimization algorithm and resampling method for predicting the swelling rate of irradiated type 316 stainless steels

Abstract

This study introduces innovative hybrid machine learning (ML) models that integrate seven state-of-the-art ML algorithms with the WEighted Relevance-based Combination Strategy (WERCS) and the Fire Hawks Optimizer (FHO) algorithm to predict the swelling rate of irradiated type 316 stainless steel (316 SS). A database of 333 experimental samples with 19 selected input features is utilized for model development. The WERCS method is used to address dataset limitations related to size and imbalance, while hyperparameter optimization is efficiently performed using cross-validation combined with the FHO algorithm. Performance evaluation across multiple metrics identifies the WERCS-FHOCGB model, which combines WERCS, FHO, and categorical gradient boosting (CGB), as the most accurate for swelling rate prediction. To enhance interpretability, the Shapley Additive Explanations method is applied to analyze the global and local contributions of input variables, highlighting irradiation fluence, pre-irradiation fluence, dislocation density, temperature, and Si (wt.%) as the most influential factors. Additionally, the impact of these key parameters on the swelling rate of irradiated 316 SS is thoroughly investigated. Finally, a user-friendly graphical interface tool and web application are developed based on the WERCS-FHO-CGB model, providing a practical and cost-effective solution for predicting the swelling rate of irradiated 316 SS.