Enhancing Heart Disease Prediction with Multilayer Perceptron and Improved Blue Whale Optimization Algorithm

Abstract

Existing predictive models for heart disease predominantly concentrate on feature selection, often overlooking the importance of hyper parameter selection. To bridge this gap, our research introduces an enhanced Heart Disease Prediction System, coupling a Multilayer Perceptron (MLP) model with an improved Blue Whale Optimization (BWO) algorithm for efficient feature selection and hyper parameter tuning. This study capitalizes on the improved BWO algorithm, a nature-inspired optimization technique mimicking the feeding behaviour of blue whales, for feature selection. Our aim is to discern the most informative features from the available dataset, enabling a more precise and efficient prediction of heart disease. Simultaneously, we recognize the significance of hyper parameter tuning in optimizing the MLP model's performance. Hyper parameters, such as the number of hidden layers, neurons in each layer, and learning rates, greatly influence model performance, yet are not learned during the training process. To address this, our system employs the improved BWO algorithm for hyper parameter tuning, automating the search for the optimal combination that maximizes the MLP model's predictive accuracy. The integration of the improved BWO algorithm for both feature selection and hyper parameter tuning allows our Heart Disease Prediction System to enhance the accuracy and efficiency of the MLP model. Our results demonstrate improved performance in the identification and prediction of heart disease, thus, potentially contributing to early detection and intervention. This research underscores the potential of nature-inspired algorithms in enhancing the performance and efficiency of disease prediction systems.