Enhancing ECG-based authentication systems using VGG16 model and transfer learning

Eshaghi, Narges; Habibi, Mohammad; Bagheri, Nasour; Khatibi, Ali

doi:10.22061/jdma.2025.12571.1169

Enhancing ECG-based authentication systems using VGG16 model and transfer learning

Document Type : Full Length Article

Authors

¹ Department of Mathematics, Tafresh University, Tafresh 39518-79611, Iran

² Electrical Engineering Department, Shahid Rajaee Teacher Training University, Tehran 16788-15811, Iran

³ Department of Mechanical Engineering, Tafresh University, Tafresh 39518-79611, Iran

10.22061/jdma.2025.12571.1169

Abstract

This study explores a novel authentication algorithm leveraging ECG signals and deep learning models, specifically VGG16, enhanced with transfer learning. Authentication systems were evaluated based on preparation time, response time, and accuracy, with biometric data utilized to increase security. Traditional deep learning models face challenges in retraining time when data changes, prompting the proposed algorithm to incorporate new users or modify access efficiently using transfer learning. Key findings included training the VGG16 model on ECG data from 48 individuals (MITDB dataset) with a 99.45% accuracy and 120 seconds average training time. The transfer learning approach enabled adding or removing user data by adjusting SoftMax coefficients, reducing training time significantly to about 12 seconds per user with accuracy exceeding 99%. Removing a user followed a similar process with comparable results. Overall, the algorithm reduced retraining time by 72.89% while maintaining over 99.16% accuracy. Additionally, the system's response time for a new user was 37 milliseconds, demonstrating practicality for real-time applications. The study highlights the proposed algorithm's efficiency in managing user data changes while ensuring high accuracy and reduced retraining time, making it a robust solution for modern authentication systems.

Keywords