Key points are not available for this paper at this time.
Early detection and forecasting of depressive disorders continue to be crucial in the quickly changing field of psychological diagnostics. This study explores the possibility of ML and deep learning techniques to improve the accuracy of the diagnosis of depression beginning. We used a comprehensive dataset comprising patients' demographics, clinical attributes, and actigraph data to compare various ML models, including DNN, Logistic Regression, Decision Trees, Random Forests, SVM, KNN, and Gradient Boosting. The dataset, encompassing 55 participants, captures attributes ranging from Gender, Age, and education to more clinical metrics like the Montgomery-Åsberg Depression Rating Scale (MADRS) scores at two distinct time points. Our primary objective was to discern the model that offers the most balanced performance in predicting the onset of depression, especially given the dataset's challenges associated with class imbalances. Our findings underscore the prowess of Deep Neural Networks in achieving the highest overall accuracy. However, the gradient-boosting model emerged as a more balanced contender, especially in detecting the more challenging class indicative of depression onset. While all models exhibited commendable precision for the negative class, the recall for the positive class, representing actual cases of depression, posed challenges. This research underscores the potential of machine learning in revolutionizing depression diagnostics. It also highlights the importance of model selection based on the specific challenges and requirements of the dataset rather than solely relying on overall accuracy. As artificial intelligence continues to intersect with mental health, studies like these pave the way for more personalized, accurate, and early interventions, potentially transforming the prognosis for millions affected by depression globally.
Kanchapogu et al. (Fri,) studied this question.