D-t-SNE: Predicting heart disease based on hyper parameter tuned MLP

Abstract

Heart disease has recently become a major cause of high mortality rates. Concurrently, data mining (DM) has also attracted increasing attention in the healthcare field. Identifying this disease in the starting stage helps to minimize treatment costs, thereby saving people’s lives. Although several classification models have been applied in recent years, they are deficient in their prediction accuracy. Hence, this research intends to apply DM methods for heart disease prediction by concentrating on maximum accuracy. The proposed scheme is evaluated for the performances in terms of various performance metrics using HD datasets (Statlog + Hungary + Cleveland + Switzerland + long beach VA datasets). Deep Convolutional Neural Network (CNN) models have been proposed to extract relevant features owing to their capability for automatic and effective learning. Subsequently, the fusion was performed. Following this, D-t-SNE (Distributed-t-Stochastic Neighborhood Embedding) is introduced to reduce dimensionality reduction to solve over fitting issues and remove redundant data to improve the classifier performance for predicting heart disease. Furthermore, efficient classification is undertaken by the introduced hyper-parameter-tuned MLP (H-MLP), as it has the ability to solve classification issues. Finally, the proposed work was assessed through comparison with traditional techniques with respect to accuracy, precision, sensitivity, Matthew’s correlation coefficient (MCC), F1-score, specificity, and negative predictive value (NPV). The outcomes showed the superior prediction of this system compared to conventional research.