Healthcare prognosis: GradiLearn-driven elitist genetic algorithm for disease predictions

Abstract

Hybridization in evolutionary algorithms is gaining traction, boosting convergence speed and solution accuracy-a pivotal research focus. This paper introduces ensemble of GradiLearn with Non-Dominated Sorting Genetic Algorithm-II designed to optimize the neural networks with a focus on dual objectives: enhancing accuracy and minimizing Mean squared error. This paper implements GradiLearn, a robust approach of back propagation with self adaptive learning learning rate. The central concept of the proposed framework involves initiating the population through GradiLearn, rather than relying on random selection. The GradiLearn serves as a tool to optimize weights in cases where weight represents an volatile population parameter. Subsequent to population creation, it evolves with NSGA-II method, namely GRL-NSGA II to produce better generation. The efficacy of GRL-NSGA II is elevated through the enhancement of individuals within the population. This article also implements a non cooperative fitness function for the finest measure called as Accurate Classification Rate (ACR)and Canberra distance-based crowding distance, providing an absolute measure of distance. Experimental results highlight the proposed method’s effectiveness in addressing binary and multi-class classification challenges, particularly with imbalanced medical datasets. Through empirical demonstration, the article establishes the models competence in reducing neural network topology while enhancing generalization performance. Comparative analysis with various machine learning models and ensemble methods reinforces the proposed method as a robust classifier, enhancing classification process ability.