Evolutionarily optimized gradient boosting machine to accurately forecast coke reactivity index

In this study, advanced statistical and machine learning methodologies are applied to predict and analyze the Coke Reactivity Index (CRI) using an industrial dataset comprising 636 coal samples with diverse physiochemical parameters including volatile matter, sulfur, moisture, ash content, basicity index, and plastic layer thickness. Gradient Boosting hybrid models are employed with nine optimization algorithms integrated with k-fold cross-validation to tuned hyperparameter and evaluate predictive performance. The statistical performance of each optimization algorithm is rigorously assessed through metrics including R-squared, Mean Squared Error (MSE), and Average Absolute Relative Error (AARE%), revealing that Artificial Bee Colony and Whale Optimization algorithms demonstrate superior consistency and precision. Furthermore, the SHAP (Shapley Additive Explanations) analysis identifies volatile matter content as the most influential variable affecting CRI predictions, providing actionable insights into the coke-making process. The findings underscore the potential of hybridized machine learning algorithms for optimizing coke production processes.