Comparative Study of Genetic Algorithm and Artificial Neural Network for Multi-class Classification based on Type-2 Diabetes Treatment Recommendation model

Abstract

Multi-class Classification is often used for classification and categorization purposes under Machine Learning wherein vast datasets can be classified into multiple labels/classes. It is often perceived as more complex than binary classification and is still being explored and studied. The main objective of this paper is to perform a comparative study of Genetic Algorithm and Artificial Neural Network to identify the algorithm that enhances the accuracy of multi-class classification. The experimental results obtained in the comparative study are evaluated using our model developed for Type-2 Diabetes Individualistic Treatment Recommendation, which successfully implements multiclass classification of patients into 7 classes(Treatment Line). Presently, doctors prescribe drugs by using their knowledge and experience, but they require a faster and more efficient system to assist them in taking the final decision by providing a suitable suggestion about the treatment line. The dataset used by our model consists of 24 input attributes and 7 output class of 2430 individuals having different characteristics like hypertension etc to make it as diverse as possible. While comparing the benefits and drawbacks of these two algorithms on our model, we have considered factors such as accuracy, training, testing and complexity. Among the two types of classifier the ANN classifier leverages the performance of the system by giving the most accurate result and generating the prediction accuracy of 92%. Thus, based on the comparative study ANN classifier demonstrates better prediction results than evolutionary Genetic Algorithm.