RESUMO
Computer-aided diagnosis (CAD) systems play a vital role in modern research by effectively minimizing both time and costs. These systems support healthcare professionals like radiologists in their decision-making process by efficiently detecting abnormalities as well as offering accurate and dependable information. These systems heavily depend on the efficient selection of features to accurately categorize high-dimensional biological data. These features can subsequently assist in the diagnosis of related medical conditions. The task of identifying patterns in biomedical data can be quite challenging due to the presence of numerous irrelevant or redundant features. Therefore, it is crucial to propose and then utilize a feature selection (FS) process in order to eliminate these features. The primary goal of FS approaches is to improve the accuracy of classification by eliminating features that are irrelevant or less informative. The FS phase plays a critical role in attaining optimal results in machine learning (ML)-driven CAD systems. The effectiveness of ML models can be significantly enhanced by incorporating efficient features during the training phase. This empirical study presents a methodology for the classification of biomedical data using the FS technique. The proposed approach incorporates three soft computing-based optimization algorithms, namely Teaching Learning-Based Optimization (TLBO), Elephant Herding Optimization (EHO), and a proposed hybrid algorithm of these two. These algorithms were previously employed; however, their effectiveness in addressing FS issues in predicting human diseases has not been investigated. The following evaluation focuses on the categorization of benign and malignant tumours using the publicly available Wisconsin Diagnostic Breast Cancer (WDBC) benchmark dataset. The five-fold cross-validation technique is employed to mitigate the risk of over-fitting. The evaluation of the proposed approach's proficiency is determined based on several metrics, including sensitivity, specificity, precision, accuracy, area under the receiver-operating characteristic curve (AUC), and F1-score. The best value of accuracy computed through the suggested approach is 97.96%. The proposed clinical decision support system demonstrates a highly favourable classification performance outcome, making it a valuable tool for medical practitioners to utilize as a secondary opinion and reducing the overburden of expert medical practitioners.