International Journal on Advanced Science, Engineering and Information Technology, Vol. 13 (2023) No. 2, pages: 578-584, DOI:10.18517/ijaseit.13.2.17701

Hybrid Feature Extraction and Infinite Feature Selection based Diagnosis for Cardiovascular Disease Related to Smoking Habit

Umera Banu, Kalpana Vanjerkhede

Abstract

Electrocardiography (ECG) is a growing study in the realm of patient monitoring systems to detect cardiovascular disease (CVD) by smoking habits. This study investigated the categorization and analysis of CVD related to smoking habits using the ECG dataset from the Physikalisch-Technische Bundesanstalt (PTB). After acquiring ECG data, the feature vectors were extracted using hybrid feature extraction (a mix of statistical, energy, and entropy characteristics). To extract features from obtained ECG signals, nineteen characteristics were merged. Artifacts in the signal are being reduced by using a zero phase butterworth filter, and the peak identification of ECG signal is attained by using the Pom-Tompkins method. Then, infinite feature selection was used to delete unnecessary characteristics or choose the best feature subsets. After choosing the best characteristics, the ECG signals of smokers and non-smokers are classified using a supervised classifier (K-Nearest Neighbor (KNN)). KNN classifier has the advantage of balancing the data for the classification of smoker and non-smokers. This discovery has several benefits, including earlier detection of cardiovascular disorders and great assistance to physicians during surgery. The results of the experiment are evaluated using classification Accuracy, F-Score, Specificity, Sensitivity, and Mathews Correlation coefficient (MCC) for the proposed technique, and the process efficiently discriminated the ECG signals of smokers from non-smokers in comparison to the previous methods; the suggested strategy improved accuracy by 3-40%.

Keywords:

Electrocardiography; hybrid feature extraction; infinite feature selection; k-nearest neighbor

Viewed: 110 times (since abstract online)

cite this paper     download