Predicting the Depression based on Motor Activity of Individuals using Time Series Forest Classifier

Depression is a prevalent and perilous mental health condition impacting a significant portion of the global population. It manifests as prolonged sadness, categorized as unipolar, or as one of the phases of bipolar disorder, profoundly affecting individuals and their families. A considerable number of suicides worldwide stem from depression, underscoring the urgency of early detection and symptom prediction for societal well-being. Traditional psychiatric methods exist for diagnosis, but AI advancements offer promising avenues for pre-emptive intervention. This paper introduces a pioneering approach utilizing motor activity patterns of depression patients for classification and prediction. Changes in daily activity signify the onset of depressive episodes, data which can be gathered through various devices like smartwatches and fitness apps. Leveraging time series forest classifiers, the proposed model achieves a prediction accuracy of 77.7%. However, for a healthcare problems, specificity and sensitivity are better metrics compared to accuracy, as they throw light on how good the model is in rightly identifying patients and non-patients. The model demonstrates an AUC of 85% on the ROC curve, indicating its efficacy in patient identification.