An enhanced grey wolf optimization method for feature selection and explainable prediction in chronic disease analytics

The advent of artificial intelligence paradigms has contributed a great deal towards improving predictive accuracy in medicine especially in the prompt diagnosis of chronic diseases. Feature selection is crucial in refining classification models by identifying and eliminating non-informative or repetitive data attributes. In the process of feature selection, the classical Grey Wolf Optimizer (GWO) suffers from low convergence efficiency and likelihood of stagnation at local optima. This research explores the median random wandering fitness-based Grey Wolf Optimization (MRWF-GWO) method to enhance local search performance and obtain an optimal balance between exploration and exploitation. The effectiveness of the proposed approach was validated through a systematic analysis using classification metrics across ten chronic disease datasets. Additionally, the performance of MRWF-GWO is compared with several state-of-the-art metaheuristic algorithms. The experimental analysis substantiates that MRWF-GWO consistently achieves superior performance in terms of search efficiency, classification accuracy, feature subset size, stability, computational time, and convergence rate. MRWF-GWO yielded an average accuracy of 92.89% across the ten chronic disease datasets considered in the study. The MRWF-GWO algorithm achieved a substantial feature size reduction (95%-99%) compared to the original set thereby demonstrating consistent efficiency in feature selection. Additionally for Leukemia and Lung datasets, the proposed MRWF-GWO method has the fastest execution time of 4.33s and 6.03s compared to all other metaheuristic optimization algorithms (MHA). The outcome of the predictive model is used to comprehend the most relevant clinical features that contributed to predicting chronic diseases, using an explainable machine learning technique known as Shapley Additive Explanations.