Detection of Schizophrenia from Actigraphy Data using UMAP Features

Schizophrenia is a prevalent public health problem that costs both health and the economy. In most cases, there is no straightforward quantitative test for diagnosing mental health issues. Conventional methods based on patient interviews or questionnaires are still used for diagnosing mental health problems like depression and schizophrenia. This paper proposes a new approach to diagnose schizophrenia and related mental health problems by using motor activity data. The proposed framework consists of two parts: feature extraction using UMAP (Uniform Manifold Approximation and Projection) and a Neural network (NN) for classification. The results from using this framework on 24-hour motor activity data (Psykose Dataset) show high accuracy in diagnosing schizophrenia with a score of 93.89%, sensitivity of 89.04%, and specificity of 96.42%. This new approach demonstrates improved results compared to traditional methods of manual feature extraction.