Multiple-criteria decision making, feature selection, and deep learning: A golden triangle for heart disease identification

Abstract

Cardiovascular diseases (CVD) are among the deadliest illnesses that suffer many people worldwide. Nevertheless, an on-time diagnosis of heart disease can play a significant role in healthcare; since it can reduce the chances of death and save much money in case of necessary treatment. This article develops an efficient and accurate system that uses artificial neural networks (ANN), feature selection (FS) methods, and multiple-criteria decision-making (MCDM) techniques to diagnose heart disease. Appropriate features are selected by utilizing five feature selection methods. Then, three artificial neural networks for CVD prediction were implemented, and their performances were checked according to eight solid criteria. In addition, Particle Swarm Optimizer (PSO) has been utilized to select the optimal hyperparameters to reach high accuracy. A novel integrated weighting method using MEREC and BWM is developed to consider experts’ statements and data evaluation in the criteria weighting operation to choose the best feature selection-artificial neural network method for heart disease detection. Then, an extended version of ELECTRE III is introduced to assess fifteen alternatives, which has proven to be more accurate than other ELECTRE methods in this research. Results validation has been done by utilizing eight MCDM techniques, the Pearson correlation coefficient test, and criteria weights sensitivity analysis by implementing ten scenarios to assess our methodology’s robustness. Finally, the outcomes indicate that LASSO-CNN, AdaBoost-CNN, and AdaBoost-MLP are the top 3 approaches with the accuracy of 99.51%, 98.54%, and 98.54%, respectively. To the best of our knowledge, these are the highest accuracy rates obtained in the literature so far.