Explainable artificial intelligence for multiclass prediction model of suicide attempt

Abstract

Suicide attempt prediction is a challenging classification problem that involves a variety of risk factors in individuals with various medical conditions. Accurate risk stratification prediction is hampered by the absence of reasons for those who have attempted suicide and developing prediction model is challenged to be explained. Therefore, this work aimed to develop a multiclass prediction model for suicide attempts and to use Shapley additive explanations (SHAP), an explainable artificial intelligence (XAI) method to analyze the prediction model for suicide attempts in explaining the decision of the model. The prediction model is trained using machine learning approaches, random forest (RF) and gradient boosting (GB), on a clinical dataset of patients with chronic diseases. GB demonstrated higher accuracy with 0.81 than RF with 0.78 for multiclass classification results (no risk, low risk, moderate risk, and high risk). By analyzing the SHAP explanations, clinicians can gain valuable insights into the factors contributing to suicide attempt predictions in patients with chronic diseases. This enhanced understanding can facilitate more informed and appropriate treatment decisions, potentially leading to improved patient outcomes and targeted interventions.