Classification of ECG Signals Using Machine Learning Techniques

Abstract

Cardiovascular diseases are one of the leading causes of mortality in contemporary society. With the growth in the accumulation of medical data, new opportunities have arisen to enhance diagnostic accuracy using machine learning techniques. Heart diseases present symptoms that can be similar to other disorders or be mistaken for signs of aging. Furthermore, diagnosing based on electrocardiogram (ECG) signals can be challenging due to the variability in signal length and characteristics. This article has developed a methodology for classifying ECG signals using the k-Nearest Neighbor (kNN) algorithm and statistical techniques. 9000 ECG signal samples from the PhysioNet database were processed. The signals were normalized to a length of 9000 samples, and relevant features for classification, such as median, standard deviation, skewness, among others, were extracted. Multiple kNN models with different parameters were trained and evaluated on a test set. The models exhibited high performance in classifying normal signals but faced difficulties in correctly classifying signals with arrhythmias. The weighted kNN algorithm demonstrated the best accuracy, although all models showed a tendency to misclassify abnormal signals due to data imbalance. While significant accuracy was achieved in ECG signal classification, there is still room for improvement. Future strategies could involve extracting more relevant features, addressing data imbalance, and fine-tuning model hyperparameters. Integrating domain knowledge from the medical field and advanced signal processing techniques could further enhance classification accuracy.

Received: 3 January 2024 / Accepted: 7 April 2024 / Published: 5 May 2024