Hybrid Feature Selection with Chaotic Rat Swarm Optimization-Based Convolutional Neural

Abstract

Introduction: Early diagnosis of Cardiovascular Disease (CVD) is vital in reducing mortality rates. Artificial intelligence and machine learning algorithms have increased the CVD prediction capability of clinical decision support systems. However, the shallow feature learning in machine learning and incompetent feature selection methods still pose a greater challenge. Consequently, deep learning algorithms are needed to improvise the CVD prediction frameworks. Methods: This paper proposes an advanced CDSS for CVD detection using a hybrid DL method. Initially, the Improved Hierarchical Density-based Spatial Clustering of Applications with Noise (IHDBSCAN), Adaptive Class Median-based Missing Value Imputation (ACMMVI) and Clustering Using Representatives-Adaptive Synthetic Sampling (CURE-ADASYN) approaches are introduced in the pre-processing stage for enhancing the input quality by solving the problems of outliers, missing values and class imbalance, respectively. Then, the features are extracted, and optimal feature subsets are selected using the hybrid model of Information gain with Improved Owl Optimization algorithm (IG-IOOA), where OOA is improved by enhancing the search functions of the local search process. These selected features are fed to the proposed Chaotic Rat Swarm Optimization-based Convolutional Neural Networks (CRSO-CNN) classifier model for detecting heart disease. Results: Four UCI datasets are used to validate the proposed framework, and the results showed that the OOA-DLSO-ELM-based approach provides better heart disease prediction with high accuracy of 97,57 %, 97,32 %, 96,254 % and 97,37 % for the four datasets. Conclusions: Therefore, this proposed CRSO-CNN model improves the heart disease classification with reduced time complexity for all four UCI datasets