Using Machine Learning of Demographics and Clinical History to Predict Suicide Outcomes

According to the CDC, there were 47,511 suicide deaths in the US in 2019 alone. This excludes suicide attempts and cases not reported as suicide due to social stigma. Predicting who attempts or commits suicide is challenging, but recent advances in machine learning may allow for improvements. In this project, mood disorder and related clinical features were extracted from free-text descriptions to predict suicide mortality. Using these features and demographic data on deceased individuals collected by the National Institute of Mental Health, a machine learning model was trained to predict suicide/non-suicide outcome with a weighted F-1 score of ~84%.