An examination of the hybrid meta-heuristic machine learning algorithms for early diagnosis of type II diabetes using big data feature selection

Abstract

People are increasingly getting type II diabetes mellitus (T2DM) due to unhealthy food styles, decreased outdoor activities caused by the COVID-19 pandemic, and unawareness of type II diabetes risk factors. This disease is hidden in the early stages and causes many comorbidities like fatty liver, heart disease, and peripheral artery disease. This study presents several hybrid algorithms to diagnose T2DM in its early stages without requiring expensive and time-consuming medical tests. We first apply feature selection using the Particle Swarm Optimization (PSO) algorithm to reduce the required computations. Meta-heuristics are used in developed hierarchical algorithms to optimize the hyperparameters of machine learning algorithms for classification. A comparative analysis of the algorithms with performance metrics shows Genetic Algorithm-Support Vector Machine (GA-SVM) has the largest area under the Receiver Operating Characteristic (ROC) curve (0.934) and better performance in most metrics (Accuracy of 0.934 and F1-Measure of 0.945) and reasonable metaheuristic computational time. Therefore, the GA-SVM algorithm is recommended for clinical decision support systems. This algorithm diagnoses T2DM at early stages by responding to several questions with about 93% accuracy, which can help patients to survive future complications through lifestyle intervention therapy.