Roya Zare Hanji, Ali Mahdizade, Amin Zare Hanji

Abstract

A large number of people perish or their functions are affected by heart disease in the world every year. The death toll from heart problems is much higher than other accidents or natural disasters. This is the main reason for the development of scientific and research activities in medical sciences and in the context of heart disease. It is widely and rapidly expanded into other sciences, such as engineering, to find effective preventive methods for these diseases. In this study, such an objective has been developed with the aim of helping the development of cardiac arrhythmia detection algorithms using data from measurement systems of cardiac characteristics, such as electrocardiogram signals. The features were extracted from the pulses according to the shape of the captured signals and the changes in the process. Then, the frequency and temporal properties of the signal were studied simultaneously by examining different methods, among which the UWMV-GSA method was selected for this purpose. After feature extraction, the feature length was reduced in the next step by applying the PCA operator to reduce the feature space dimension. In the final step, the received signals were classified based on the extracted feature vectors through combining information in the classification of cardiac arrhythmias using five base classifiers of LMS, RLS, decision tree, KNN, and SVM to classify cardiac datasets. Furthermore, the weight was selected based on the gravitational search algorithm (GSA) optimization. Different hybrid classification methods are compared in this study and a new UWMV-GSA algorithm is proposed for dynamic dataset classification in which each specialist (base classifier) ​​ predicts a specific area of ​​the dataset with the highest accuracy. Thus, system performance is increased by up to 90% by combining the results. The accuracy and weight of each classifier and the UWMV-GSA for test and training data are presented in various tables and graphs.