An End-to-End Model for Mental Disorders Detection by Spontaneous Physical Activity Data

Mental disorders cannot only bring tremendous burdens to patients themselves, but also to the society. Effective early prediction and symptom monitoring can significantly improve mental health care across different populations. In this aspect, research on detecting mental disorders based on spontaneous physical activity (SPA) data has yielded promising results. However, when using SPA data, traditional methods of manually extracting features require highly specialised knowledge in signal processing. This has made the development of this research in the field of mental health extremely challenging. To this end, we propose an end-to-end method based on SPA data to address the challenges of time-consuming manual feature engineering and high requirements for domain expertise. The end-to-end approach allows researchers to focus solely on data and results, which is of significant importance for detecting, and real-time monitoring mental health using sensor data from wearable devices like SPA. We take a long-short term memory (LSTM) model with embedding layers for classification. Experimental results have demonstrated that, the end-to-end method is effective in detecting diseases with a binary classification task. The unweighted average recall (UAR) on the test set of the classification tasks shows that this model bears significant effectiveness in tasks related to detecting health conditions or diseases. In the multi-class task of disease detection, the results indicate that further research is needed on the data features of different diseases.