Enhancing Suicide Detection via Chi-Square-Based Feature Selection and Machine Learning

Abstract

Suicidal ideation detection has emerged as a crucial issue in mental health research, especially as social media use has increased and users are more likely to reveal psychological distress. In this study, an AI-driven approach to detecting suicide intent on Twitter is proposed using machine learning techniques. Natural Language Processing (NLP) techniques were used to collect and preprocess a dataset of tweets, from which features were selected using the SelectKBest algorithm, which is based on chi-square analysis. Using an 80:20 train-test split, four classification models—AdaBoost, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Decision Tree—were trained and evaluated. The SVM classifier achieved the highest accuracy of 89.66% and the highest precision of 91%, outperforming the other classifiers in detecting suicidal tendencies in social media posts. For the early detection of suicide risk, the results validate the effectiveness of conventional machine learning models when combined with appropriate feature selection techniques. This work shows how AI can be integrated into real-time mental health monitoring systems for early intervention and prevention.