فهرست نشریات علمی دسترسی آزاد ایران

In today’s rapidly shifting healthcare landscape, proper diagnosis is critical for efficient illness management and treatment. Diagnostic tests give critical information for determining the underlying cause of symptoms, allowing for early and focused medical treatments. Therefore, this work suggests a hybrid multi-objective evolutionary feature selection (FS) mechanism based on the CrayFish Optimization Algorithm (COA) and Harris Hawk Optimization (HHO) with a novel mutual information-based initialization for identifying the actual cause of a particular disease. This paper also proposes a novel classifier ensemble technique using the Choquet fuzzy integral, with basic classifiers based on Gaussian Naive Bayes (GNB), Support Vector Machine (SVM), Multi-layer Perceptron (MLP), and Functional Link Artificial Neural Network (FLANN) models for accurate classification of the disease by using the selected features. For experimental purposes, we have taken five disease datasets: Survey Lung Cancer, Diabetes, Cardiovascular, Heart, and Heart1 from the UCI repository. More specifically, the ensemble approach achieves 92% accuracy for the Survey Lung Cancer dataset, outperforming SVM and C-FLANN, both with 90%. In the diabetes dataset, too, the ensemble classifier boosts performance to 91%, surpassing P-FLANN, with 88%. The cardiovascular dataset shows a similar pattern whereby the ensemble model achieves 89% accuracy, surpassing the 86% attained by P-FLANN. While in the Heart1 dataset it reaches 95%, exceeding MLP’s 93%, the ensemble classifier only somewhat increases accuracy to 95% in the Heart dataset from MLP’s 94%. The suggested ensemble approach performs up to 2% better than conventional ensemble methods, including stacking, weighted averaging, and majority voting. These findings unequivocally show how much better the suggested ensemble method performs consistently over several datasets. By combining strong classifiers and selecting the best features, the suggested method outperforms separate models, making it suitable for disease diagnosis.