Predictive Modeling for Depression Diagnosis Using Machine Learning and DSM-5 Criteria

Depression is a pervasive mental health disorder that significantly impacts individuals’ quality of life and global productivity. Early and accurate diagnosis is essential for effective treatment, yet the process is often hindered by stigma and limited access to mental health professionals. This study presents a machine-learning-based predictive model for diagnosing depression using a dataset labeled according to DSM-5 criteria. The dataset, comprising 363 instances and 23 attributes across diverse demographics, was collected through an online survey and labeled with guidance from a professional psychiatrist. After preprocessing, we explored nine machine learning algorithms, focusing on balancing accuracy, precision, recall, and F1 score to identify the most effective approach. AdaBoost emerged as the top-performing algorithm, achieving an accuracy of 99.45% with an AUC score of 0.9. All results were validated through 10-fold cross-validation to ensure robustness and mitigate overfitting risks. To develop a personalized healthcare solution, we deployed the best model with fine-tuning in a web-based depression detection system. We anticipate that this system will enhance access to mental health facilities and raise awareness. Our findings suggest promising potential and propose future work to scale the dataset, integrate additional features such as behavioral and physiological data, and validate the model on external datasets while considering the ethical implications of real-world deployment.