A Linearly Adaptive Sine–Cosine Algorithm with Application in Deep Neural Network for Feature Optimization in Arrhythmia Classification using ECG Signals

Abstract

This work proposes a novel variant namely Linear Adaptive Sine–Cosine Algorithm (LA-SCA) of newly developed metaheuristic called Sine–Cosine Algorithm (SCA). The developed variant utilizes a linearly adaptive operator that is based on the number of generations, followed by an Opposition based Learning (OBL) model applied around the best solution. The applied mechanisms help LA-SCA to overcome the slow convergence problem in SCA and capable of achieving appropriate balance between macro and micro search ability of SCA. The performance of proposed LA-SCA is evaluated using a set of 23 state-of-the-art benchmark problems having a variety of characteristics and diverse complexities. The LA-SCA is statistically analysed by applying a Wilcoxon ranksum test and performance analysis using convergence curves and diversity analysis. In second part, the proposed LA-SCA is utilized to offer an intelligent Deep Neural Network (DNN) mechanism for improving the feature extraction of ECG signal for efficient classification of arrhythmia diseases. During the pre-processing step, Discrete Wavelet Transformation (DWT) is utilized to categorize ECG data followed by improving the extraction of QRS complex features using proposed LA-SCA. The MIT-BIH Arrhythmia ECG dataset is used in order to perform classification of signals among sixteen categories of arrhythmia disease. By using the classification analysis on the basis of variation including optimized feature vector and the LA-SCA method with DNN, the heart rate has been observed with an improved 99.29 percent accuracy as compared to other techniques. This achieved accuracy demonstrated the superiority of proposed model for appropriately classifying the ECG signal.