Utilizing Metaheuristic Machine Learning Techniques for Early Diabetes Detection

Abstract

The pervasiveness of diabetes mellitus has been raised quickly in recent years. According to present statistics nearly 90% of diabetic people have type 2 diabetes globally. The only solution for this is an early prevention strategy, as it has no longterm proper treatment. Nowadays, Metaheuristic optimization algorithms along with various deep learning and machine learning algorithms and techniques are now being utilized in the medical field for chronic diseases with effectiveness and better outcomes than conventional methods. In this paper the authors objective is to increase the effectiveness of early diabetes detection and suggesting the food items based on carbohydrates, fiber and sugar content. The Bat metaheuristic approach is inspired by biological systems, was utilized to choose the key features during the preprocessing of medical data. The performance of the suggested methods was contrasted with that of BAT nature optimized algorithm, which is metaheuristic bio-inspired feature selection algorithm. Different classifiers were then used in conjunction with the BAT and Firefly algorithms. Then datasets used for implementation is Pima India Diabetes Dataset, Hospital Frankfurt and food nutrition data set for diet recommendation. Smote is applied for data balancing class distribution. The outcomes demonstrate that the BA approach performs better on balanced dataset than, with 97% accuracy especially when the dataset's instance count rises.