A machine learning framework for the multi-class prediction of suicide attempt risk

Suicide attempt risk stratification is often hampered by the limitations of current binary classification methods, hindering effective management of suicidal behaviour. Despite the increasing cases of suicidal behaviour in high-risk populations, the number of studies on individuals with Systematic Lupus Erythematosus (SLE) with suicidal attempts is limited and remains unclear. SLE is a chronic autoimmune disease where body's defence system gets confused and starts attacking healthy parts of body. Therefore, this study aimed to propose a framework for multi-class prediction of suicide attempt risk in SLE individuals using machine learning. Logistic regression, decision tree, support vector machine, random forest, and gradient boosting were trained and evaluated using five-fold cross-validation on a clinical dataset of 130 patients with SLE. The multi-class models achieved an overall classification performance between 0.75 and 0.85 to classify four risk levels for suicidal behaviour (no risk, low risk, moderate risk, high risk). Specifically, the gradient boosting model trained with eleven features showed the highest precision at 0.82 in predicting suicide attempts in high-risk SLE individual. Feature importance analysis indicated that past suicide attempts and depressive disorders were the strongest predictors of suicide risk in SLE patients. While the multi-class model showed promising precision, the small sample size limits generalizability and requires validation with larger populations. Nonetheless, this preliminary study highlights the potential of risk stratification for supporting early identification and improving clinical decision-making in suicide prevention.