Detecting Mental Health Issues in Students Using Limited ECG Data from Portable Devices.

Abstract

Introduction: Mental health issues are a growing concern among university students, significantly affecting their academic performance and quality of life. Recognizing stress in students under academic pressure is crucial for improving their well-being. This study aims to identify stress patterns through heart activity, which is closely correlated with mental health issues.

Methods: An experiment was designed involving 49 participants during exam time, using low-cost portable devices based on ECG sensors. The high quality of recorded data was confirmed by good average QRS correlation metrics. To enhance the dataset and address the problem of imbalanced data, a Generative Adversarial Network (GAN) was employed to generate synthetic ECG data in two scenarios: GAN 1, which synthesized the minority class only, and GAN 2, which synthesized both classes. A comprehensive set of Heart rate variability (HRV) indices from time, frequency, and non-linear domains were extracted for analysis. Finally, two 2 ensemble learning models were utilized to perform the stress recognition based on the HRV features set.

Results: Through cross-validation and random-split validation, our findings demonstrated significant improvements in model performance with the addition of synthetic data. Specifically, GAN 1 data improved recall, effectively capturing more stress instances, while GAN 2 data enhanced precision, ensuring accurate stress identification. The Random Forest model showed exceptional capability in managing class imbalance, further validating the effectiveness of our approach. Additionally, the use of a natural stressor, such as exam time, confirmed the practical applicability of our models.

Conclusion: These results underscore the potential of dataset enrichment in machine learning, particularly in health-related applications, and provide a robust foundation for future research and real-world validation of synthetic data's benefits in stress recognition tasks.