Detection of Suicide and Depression for Early Intervention and Initiative-taking Mental Healthcare

Abstract

This study presents an AI-based Suicide and Depression Detection system for early mental health intervention. Using advanced deep learning algorithms, particularly Convolutional Neural Networks (CNNs), the system aims to identify subtle facial cues that may signal mental health issues. By combining traditional techniques such as Histogram of Oriented Gradients (HOG) and Local Binary Patterns (LBP) with CNNs, the model effectively recognizes crucial facial features and textures linked to depression and suicide risk. This integration of HOG and LBP’s interpretability with CNN’s ability to learn complex patterns enhances the system’s accuracy and reliability, while also providing valuable insights into the specific features associated with mental health concerns. The system’s performance metrics demonstrate the effectiveness of this approach, with HOG and achieving an accuracy of 0.9695, precision of 0.95235, recall of 0.96123, and an F1-score of 0.9712, while LBP outperforms with an accuracy of 0.9949, precision of 0.9819, recall of 0.9901, and an F1-score of 0.9956.