Machine Learning Based Optimal Feature Selection for Pediatric Ultrasound Kidney Images Using Binary Coati Optimization

Abstract

Chronic kidney disease (CKD) one of the most dangerous illnesses. Early detection is vital for improving survival rates and underscoring the need for an intelligent classifier to differentiate between normal and abnormal kidney ultrasound images. Features extracted from an image have a significant impact on classification accuracy. In this study, we present a Binary Coati optimization algorithm (BCOA) for feature selection in CKD, which focuses on reducing the high dimensionality features extracted from ultrasound images, including GLCM, GLRLM, GLSZM, GLDM, NGTDM, and first order, by employing BCOA-S shaped and BCOA-V shaped transfer functions that convert BCOA from a continuous search space to a binary form, which helps in the selection of optimal features to improve the classification performance while reducing the feature dimensionality. The reduced feature was evaluated using six machine-learning classifiers: Random Forest, Support Vector Machine, Decision tree, K-nearest Neighbor, XG-boost, and Naïve Bayes. The efficiency of the proposed framework was assessed based on accuracy, precision, recall, specificity, f1 score and AUC curve. BCOA-V outperformed in terms of accuracy, precision, recall, specificity, F1 score and AUC curve by 99%,100%,97%,100%, 98%, and 98%, respectively. This makes it a superior choice for CKD diagnosis and is a valuable tool for feature selection in medical diagnosis.