An optimal weighting-based hybrid classifier for Children's congenital heart diseases signal processing

Abstract

Classification is one of the most prominent modeling approaches that can be successfully applied in model-based medical support systems to make more accurate diagnostic decisions. The classification literature indicates that numerous classifiers with different characteristics have been developed and frequently applied across a wide range of medical diagnostic processes. Several features of a classifier, such as accuracy, reliability, and complexity, can be considered when choosing the most appropriate classifier for modeling purposes in a medical decision support system. Among these features, classification accuracy is one of the most critical due to its significant impact on the quality and precision of medical diagnoses. However, achieving accurate results, especially in the medical domain which often contains complex and mixed patterns, is still a commonly difficult task. Hybridization is among the most popular techniques frequently used in the classification literature to enhance accuracy. In this paper, a hybrid classifier incorporating Long Short-Term Memory (LSTM), Support Vector Machine (SVM), and Convolutional Neural Network (CNN) is proposed and applied to diagnose congenital heart disease in children. The most distinguishing feature of the proposed hybrid classifier compared to existing ones is its optimal weighting algorithm. In this study, an optimal weighting algorithm is developed which, unlike previously proposed algorithms, can guarantee that the best accuracy will be achieved. The main objective of the proposed optimal weighting-based CNN-LSTM-SVM (OCLS) hybrid classifier is to simultaneously leverage the unique advantages of CNN in feature extraction from input signals, LSTM in modeling the sequential patterns of signals, SVM in classifying regular patterns, and especially the proposed weighting algorithm to optimally integrate the outputs of these components. Experimental findings based on the Children's Congenital Heart Disease benchmark dataset demonstrate that the proposed hybrid classification model outperforms its individual base classifiers. Furthermore, it delivers superior accuracy compared to several recently introduced hybrid models in the existing literature. Notably, the proposed method also outperforms CNN-LSTM-SVM combinations that use conventional weighting strategies such as Simple Average (SA), Majority Voting (MV), and metaheuristic optimization algorithms. Numerical results illustrate that the proposed hybrid classifier can, on average, improve the classification rate and diagnostic capability by 7.09 %, 13.60 %, 5.52 %, and 6.84 % compared to its individual components, other single shallow or deep statistical or intelligent classifiers, other weighting-based parallel hybrid classifiers, and other recently developed classifiers for congenital heart disease diagnosis, respectively. In addition, these improvements are not limited to the classification rate. The obtained results indicate that the proposed classifier can, on average, enhance performance by 6.46 %, 10.35 %, 8.37 %, and 6.22 % in precision, sensitivity, F1-score, and specificity, respectively. Consequently, it can be concluded that the proposed OCLS hybrid classifier can be an efficient alternative approach for the diagnosis of heart signal-based diseases.