An Effective Preprocessing Data on Performance of Machine Learning for ECG-Based Personal Authentication
Year: 2023
Type: conference-paper
Abstract: A system for person identification has been simulated utilizing electrocardiogram (ECG) signals as biometric identifiers. To automate ECG signal-based biometric recognition, machine learning algorithms are employed. Generally, the feature extraction stage is utilized to obtain the desired features from the ECG signal before it is processed by a machine learning algorithm. However, complex feature extraction can lead to prolonged calculation times. Additionally, the limited number of ECG signal records per person results in relatively small datasets that are not efficient for training machine learning algorithms. In this study, various classification algorithms, including Support Vector Machine (SVM), Random Forest, Naïve Bayes, and K-Nearest Neighbor (KNN), were employed to construct classification models. Additionally, multiple data segmentation techniques were employed to augment the dataset. The objective of the research is to determine the most effective techniques in terms of improving the performance of the classification models. The proposed segmentation technique creates an overlapped segment based on the R peak as a starting point. Employing datasets from this technique, the Random Forest classification model produces a personal identification accuracy of 85%. This technique provides a more than 40% improvement compared to the two previous segmentation techniques and exhibits a significant increase following t-tests with P values of 0.0168 and 0.0173. This research presents a method for segmenting raw ECG signals that is user-friendly and can enhance personal identification accuracy.
Institutions: Lambung Mangkurat University, Binus University
Language: English
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4390280320