Autoencoder와 SVM을 결합한 심전도 이상 탐지 연구

초록

The electrocardiogram (ECG) is a crucial medical data that records the electrical signals of the heart, enabling the early diagnosis of cardiovascular diseases. Despite the various deep learning models proposed for detecting abnormal ECG signals, their anomaly detection performance has been limited due to the class imbalance in ECG datasets. To address this issue, this study proposes an automated detection model combining Autoencoder (AE), which excels at learning and reconstructing normal data, with Support Vector Machine (SVM), which classifies abnormal signals. While AE is effective in improving anomaly detection performance by learning data complexity, it has a limitation in requiring a manually set threshold. On the other hand, SVM, widely used for anomaly detection, suffers from performance degradation due to data imbalance and complexity. In this paper, the two models are combined, with AE compensating for data complexity and SVM automatically classifying abnormal signals based on the reconstructed data. Experimental comparisons between the proposed model and the conventional SVM demonstrate that the proposed model achieves superior performance.

키워드