Cardiac signals classification based on Extra Trees model

Abstract

The challenging aspect of diagnosing cardiac abnormalities is the accurate classification of cardiac signals. Classifying cardiac abnormalities having similar wave morphologies by manual measurements is a herculean task. The main aim of this work is to classify Sinus Rhythm (SR), Sinus Tachycardia (ST) under physical stress, and Atrial Tachycardia (AT), by developing a machine learning model, Extra Trees (ET). The input feature set of the developed ET model consist of clinical morphologies of the cardiac signal. The clinical morphologies include durations (in ms) of P wave, PR Interval (PRI), QRS complex, T wave, QT Interval (QTI), PP Interval (PPI), and amplitudes (μV) of P, R, and T waves. Apart from classifying, the ET model has also ranked the essential clinical features required to diagnose SR, ST and AT signals. According to the ET model, P (ms), PPI (ms), and P (μV) are the crucial features to classify signals. The precision, recall, and F1 scores of the developed ET model in SR are 0.99, 0.929, and 0.963, respectively, in ST is 0.99 and in AT are 0.947, 0.99, and 0.973, respectively. The advantage of ET model over other classifiers is that, being an ensemble-based classifier developed from a decision tree classifier, it prevents over fitting.