Publication: Design and implementation of a lightweight model for ECG biometric authentication
Loading...
Date
2024-08
Authors
Kavin Raj a/l Kalyanasundram
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Electrocardiogram (ECG) is a revolutionary concept in authentication systems. Using ECG biometric authentication systems provides added benefits to the security system. This project focuses on designing and implementing a lightweight RNN model for ECG biometric authentication, aiming to achieve efficient performance with minimal resource utilization. The study involves augmenting and preprocessing raw ECG signals, including filtering, QRS complex detection, segmentation, and data augmentation techniques such as noise addition, time shifting, and pitch shifting. Three models were developed: a lightweight model with SimpleRNN layers, a moderately lightweight model with additional LSTM and GRU layers, and a heavyweight model incorporating multiple LSTM and GRU layers. These models were trained and evaluated using the "HeartPrint: ECG Biometric Recognition Based on Longitudinal Data" dataset. The training was conducted on a laptop with the following specifications: AMD Ryzen 9 4900HS processor with Radeon Graphics (3.00 GHz), 16.0 GB RAM, and a 64-bit Windows 10 Home Single Language operating system (version 22H2). The lightweight model demonstrated a competitive accuracy of 76.87% and an F1 score of 71.61, with a training time of 238.84 seconds, prediction time of 0.400189 seconds per sample, model size of 1.96 MB, and peak RAM usage of 67.62 MB. While the moderately lightweight model achieved the highest accuracy of 84.01% and an F1 score of 78.97, and the heavyweight model achieved an accuracy
of 77.92% and an F1 score of 72.29, the lightweight model stands out for its exceptional efficiency.