Depression Detector using ensemble model

This project presents a machine learning-based system designed to detect symptoms of depression using voice recordings. The motivation stems from the global mental health crisis, where millions remain undiagnosed due to limited resources, stigma, and lack of accessible tools. Our goal is to enable self-assessment of mental health in a secure, private, and user-friendly manner.

We utilize the DAIC-WoZ dataset, which contains audio recordings from clinical interviews. Relevant features were extracted from the audio data and refined using dimensionality reduction techniques (PCA). Three machine learning models—Logistic Regression, XGBoost, and LightGBM—were trained individually and as part of an ensemble, achieving an F1-score of 87%.

The system is deployed through Next.js web application, where users can upload their voice directly from the browser. After analysis, an accurate report will be generated using the model and saved securely on blockchain technology, ensuring privacy and data integrity.

This solution bridges machine learning and digital health to support early detection and intervention—especially in underserved or remote areas.

This research has been accepted for publication in Springer and will be available online soon.

#Next #Express #Node #Flask #Blockchain #Solidity #Smart Contracts #Mongodb