ADAM-sense: Anxiety-displaying activities recognition by motion sensors

In the field of Human Activity Recognition (HAR), human activities are recognized based on sensors’ streaming data. HAR has been utilized widely in various fields of application where the studies of human behaviors are conducted such as healthcare, personal care, aged care, and several other domains. This approach can also be beneficial in the field of psychiatry where the patients suffer from mental, emotional, and behavioral disorder. According to American Psychiatric Association (APA), the most common form of mental disorder is Anxiety Disorder (AD) effecting 30% of the adult population at some point in their life. In this paper, a HAR based method is proposed to recognize some behaviors pertaining to anxiety display. To make such model, a novel dataset of anxious behaviors is also created with unique features using motion sensors of smartphone and Inertial Measurement Unit (IMU). Several deep learning-based models are created and compared against random forest and gradient boost algorithms, where a deep model comprising Convolution Neural Network (CNN) and Long-Short Term Memory (LSTM) is shown to perform better than other models and could recognize anxiety-related behaviors with over 92% accuracy.