An Electrocardiogram Signal Preprocessing Strategy in the LSTM Algorithm for Biometric Recognition

Abstract

Electrocardiogram (ECG) signals are a very important tool for clinical diagnosis and can be used as a new biometric modality. The aim of this research is to determine the results of ECG signal processing using RNN methods such as the Long Short Term Memory (LSTM) algorithm by utilizing several preprocessing techniques. In this study, the ECG signal itself was previously tested by carrying out the LSTM classification process without preprocessing, and the results obtained were 0% accurate, so preprocessing was needed. The preprocessing methods tested with the LSTM classification method are Adjacent Segmentation and R Peak Segmentation to find out which preprocessing techniques greatly influence LSTM classification accuracy. The experimental results were that LSTM classification with R Peak Segmentation preprocessing obtained the highest accuracy on the two data used, namely filtered and raw data, with 80.7% and 78.95%, respectively. Meanwhile, the accuracy obtained from LSTM classification when using Adjacent Segmentation preprocessing is not good. This research compares LSTM accuracy from each preprocessing stage to determine which combination has the best results in the ECG data classification process. This research also offers new insights into the preprocessing stages that can be carried out on ECG data.

Keywords

• computer science
• machine learning
• Biometric
• Electrocardiogram
• Adjacent Segmentation
• R Peak Segmentation
• LSTM