Evaluation of 'best' machine learning algorithm in classification of arrhythmia

• Main Author: Low, Jonathan Jun Zhee
• Other Authors: Ng Yin Kwee
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Bioengineering
• Online Access: https://hdl.handle.net/10356/149465
• Institution: Nanyang Technological University

Arrhythmia is abnormality in the cardiac conduction system or irregular heartbeats. For many years, professionals such as doctors have been relying on manual calculation or measurements of the electrocardiograms (ECG) graphs to classify and provide diagnosis to patients. If any anomalies are found, these patients are usually sent for further check-ups to check for any underlying, serious conditions. For many years, data scientist and bioengineers have been applying different deep learning and machine learning (ML) algorithms to ECG signals with the goal of a reliable and automated diagnosis. However, different pre-processing techniques and learning algorithms have been utilised in different occasions. Thus, the main aim of this report is to evaluate the best machine learning algorithm to be integrated into ECG systems for a fully automatic diagnosis process. Based on the current results, by using “Sym5” discrete wavelet transform and RR intervals as features, K-Nearest Neighbour, Random Forest, Artificial (Feedforward) Neural Network and Convolutional Neural Network with Gated Recurrent Unit were found to be relatively good classifiers compared to other algorithms. Based on the computational performance, K-Nearest Neighbours seemed to be the best performing algorithm among the four.