Data Augmentation Impact on Deep learning Performance for Stress Detection

The mental health of individuals has a major influence on society. Mental disorders such as depression and anxiety are related to issues and distress to function in work, social, or family gatherings. Motivated by helping people to early detect stress, we propose a deep learning model for stress detection that is trained and tested on Stress Annotated Dataset (SAD). Also, different techniques of textual data augmentation are applied to generate more data samples to study the impact on the stress detection task. The model achieves competitive results in the task of stress detection and proves that data augmentation has a great effect on the performance. As the accuracy is raised by 6.1% to 10.4 % by applying various techniques of data augmentation.