Hybrid ensemble model with optimal weightage for suicidal behavior prediction

Abstract

Suicidal behavior is a complex phenomenon that is contextually dependent and changes rapidly from one day to another. The problem in predicting suicidal behavior is identifying individuals and at-risk groups in crisis and at risk for suicide. The current predictive model, which uses machine learning techniques, has been shown to lack accuracy, and no study has attempted to use a voting ensemble model to predict suicidal behavior. The soft voting ensemble model demonstrated good performance in the healthcare setting, but assigning optimal weights for machine learning models is challenging. Therefore, this paper aims to propose a hybrid voting ensemble model to achieve optimal weights in predicting an individual with suicidal behavior. The results show that the proposed hybrid voting ensemble model can effectively classify an individual with suicidal behavior with an accuracy of 0.84 compared to other machine learning models (logistic regression, support vector machine, random forest, gradient boosting). Hybridization of soft voting with brute force algorithm has shown that the proposed hybrid ensemble model can find the optimal weights for the machine learning model in the context of predicting suicidal behavior. Furthermore, the proposed hybrid ensemble model shows that clinical data can be used to improve the performance of machine learning models in predicting an individual with suicidal behavior.