Analysis and classification of heart rate using CatBoost feature ranking model. (July 2021)

• Record Type: Journal Article
• Authors: Dhananjay, B.; Sivaraman, J.
• Is Part Of: Biomedical signal processing and control. Volume 68(2021)
• Publication Date: 2021-07
• Subjects: Attribute order -- CatBoost model -- Extra Trees model -- Prediction Value Change -- Ridge model; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102610
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An effective feature-ranking algorithm is proposed for ECG classification. CatBoost model is synthesized using the prediction value change algorithm. Input feature set consisted of clinical ECG features. The computational time of CatBoost model is minimal. The boosting algorithm present in the CatBoost classifier minimizes over-fitting. Abstract: Background: In specific contexts, it is difficult to manually differentiate Sinus Rhythm (SR), Sinus Tachycardia (ST), and Atrial Tachycardia (AT) from ECG signals. Upright P-wave is a context in which it is hard to distinguish SR, ST, and AT. Objective: The main objective of this work is to develop a machine learning model to classify SR, ST, and AT conditions from ECG. A highly-effective feature-ranking algorithm is proposed to reduce the complexity of the classification task. Methodology: A CatBoost (CB) model is used for feature ranking. The model is synthesized using the Prediction Value Change (PVC) algorithm. The ECG features, namely P-wave (ms), PRI (ms), QRS (ms), T-wave (ms), QTI (ms), P-wave (μV), R-wave (μV), and T-wave (μV), are used as the input features of the machine learning model. Results: The accuracy, sensitivity, precision, and F1 score of the CB machine learning model are 99 %, 99.17 %, 99.25 %, and 99 %, respectively. The computational time of the CB model is 0.0078 s. The Extra Trees (ET) and Ridge Classifier (RC) models were also developed, and their performances were compared with the CB model. Conclusion: The accuracy, sensitivity, precision, and F1 score of the CB model perform better than ET and RC models. The CB-based machine learning model's computational time is minimal as it uses the symmetric tree-based inference system. The boosting algorithm present in the CB classifier minimizes over-fitting issues.