Enhancing Diabetes Prediction and Classification through Metaheuristic Optimization of Deep Neural Networks

Abstract

Deep Neural Network (DNN) classifiers have found numerous applications in health care due to their ability to learn complex patterns and extract meaningful insights from medical data. These models have parameters that are not learned from the data but need to be set before the training process begins. The choice of hyperparameters can significantly impact the performance of a model. Fine-tuning these hyperparameters is a crucial step in optimizing a model for a specific task. Machine learning and metaheuristic approaches play vital roles in this process. This paper presents a hybrid metaheuristics algorithm using Coati Optimization and Cellular Automata to provide a systematic and automated approach in exploring the hyperparameter space of Deep Neural Network. To evaluate the model performance, the experiments were conducted with PIMA Diabetes dataset and a real-world data set collected from a local pathology clinic. Through stratified k-fold cross-validation, the proposed hybrid model obtains 95.73% accuracy on PIMA dataset while on the real dataset it achieves 99.9% accuracy. The strengths of our proposed model are further compared with other machine learning algorithms demonstrating its potential to revolutionize diabetes research.