Suicide Risk Prediction Using Random Forest and Imbalanced Data Management Approach

Predicting suicide risk is crucial for reducing suicide-related deaths. Machine learning techniques offer the potential for early and precise identification of individuals at risk. A primary challenge in this realm is the imbalanced nature of data, which can undermine model efficiency. In this study, a suicide risk prediction model was developed using the random forest algorithm and was compared its performance with other methods. To address data imbalance, two distinct balancing techniques were applied. Subsequently, these balanced datasets were fed into the model, and their prediction outcomes were assessed in tandem. Our findings revealed an overall prediction accuracy of 0.99, backed by a sensitivity of 0.99 and a specificity of 0.98. It’s evident that leveraging the random forest algorithm, combined with data balancing methods, significantly enhances the evaluation metrics, setting a new performance standard in predictive modeling.