Intelligent Feature Selection for ECG-Based Personal Authentication Using Deep Reinforcement Learning.

Area of Science:

• Biometrics
• Machine Learning
• Signal Processing

Background:

• Personal authentication systems require robust and efficient methods.
• Electrocardiogram (ECG) signals offer unique physiological characteristics for biometric identification.
• Optimizing feature selection is crucial for developing efficient wearable authentication devices.

Purpose of the Study:

• To investigate optimal electrocardiogram (ECG) signal features for personal authentication.
• To implement a reinforcement learning (RL) algorithm for feature selection in ECG-based biometrics.
• To enhance the efficiency and performance of wearable authentication systems.

Main Methods:

• ECG signals were collected from 11 subjects over 6 days.
• Reinforcement learning (RL) with deep learning structures was employed for feature optimization.
• Bayesian optimization hyperband was used for automatic deep learning architecture construction.
• Support vector machines (SVM) were utilized in conjunction with the optimized RL algorithm.

Main Results:

• The feature selection process is critical for improving authentication performance and reducing computational load.
• Optimized RL with SVM achieved higher accuracy (5%, 3.6%, 2.6% increase) compared to Information Gain, ReliefF, and pure RL.
• The optimized RL approach resulted in fewer selected features and generally lower equal error rates (EER).

Conclusions:

• Feature selection is essential for efficient and accurate ECG-based personal authentication.
• Reinforcement learning offers a powerful approach for optimizing feature selection in biometric systems.
• The proposed method enables the development of efficient wearable authentication devices with reduced power consumption.