Enhancing the performance of machine learning models through Meta-Heuristic Hyperparameter tuning: A comparison for breast cancer prediction

Abstract

Breast cancer is one of the leading causes of mortality and morbidity among women. Early diagnosis and timely intervention significantly improve treatment outcomes and contribute to maintaining patients' quality of life. This study investigates metaheuristic hyperparameter optimization (HPO) techniques to enhance the performance of machine learning models used for breast cancer prediction. Five different machine learning models were evaluated on a dataset consisting of 569 samples and 30 features. In addition, the performance of ten widely used metaheuristic algorithms reported in the literature was comparatively analyzed. The statistical significance of the performance differences among the algorithms was first assessed using the Kruskal–Wallis test. Subsequently, pairwise comparisons were conducted using the Mann–Whitney U test with Bonferroni correction, taking the SFOA algorithm as the reference method. The initial classification accuracy improved by approximately 2.6% following the application of metaheuristic optimization. Furthermore, a comprehensive evaluation was performed using multiple performance metrics, including accuracy, precision, recall, F1-score, confusion matrix, Cohen’s kappa, and log loss. The findings demonstrate that metaheuristic methods provide an effective and reliable approach to HPO. Moreover, while the SFOA algorithm achieved statistically significantly higher performance than some competing methods, it attained a level of predictive accuracy comparable to that of other strong-performing algorithms.