Unsupervised Transformer-Based Anomaly Detection in ECG Signals

Anomaly detection is one of the basic issues in data processing that addresses different problems healthcare sensory data. Technology has made it easier to collect large and highly variant time series data; however, complex predictive analysis models are required ensure consistency reliability. With rise size dimensionality collected data, deep learning techniques, such as autoencoder (AE), recurrent neural networks (RNN), long short-term memory (LSTM), have gained more attention recognized state-of-the-art anomaly techniques. Recently, developments transformer-based architecture been proposed an improved attention-based knowledge representation scheme. We present unsupervised method evaluate detect anomalies electrocardiogram (ECG) signals. The model comprises two parts: embedding layer a standard transformer encoder. introduce, implement, test, validate our well-known datasets: ECG5000 MIT-BIH Arrhythmia. Anomalies detected based on loss function results between real predicted ECG sequences. found use encoder alternative for enables better performance suggested remarkable ability signal outperforms approaches literature both datasets. In dataset, can with 99% accuracy, F1-score, AUC score, 98.1% recall, 100% precision. Arrhythmia achieved accuracy 89.5%, F1 score 92.3%, 93%, recall 98.2%, precision 87.1%.