Machine Learning Models for Predicting Suicidal Ideation, Depression, Anxiety, and Stress in the General Population

Teodora Matic, Aleksander Sadikov, Peter Pregelj, Polona Rus Prelog

Abstract: Suicide is a leading cause of death globally, with over 700,000 registered deaths and an estimated 25 million attempts annually. In Slovenia, suicide rates exceed the EU average, with a mortality rate of 17.08 per 100,000 in 2018. Despite extensive research, accurately predicting individuals at risk of suicide remains a challenge, with current tools showing prediction accuracy no better than chance. Depression, anxiety, and stress significantly contribute to mental health burdens, often going underdiagnosed due to overlapping symptoms and subjective screening tools. Machine learning (ML) offers promising advancements in suicide prevention by enabling data-driven predictive models. Building on our prior work predicting suicidal ideation (SI) using ML, this study incorporates depression, anxiety, and stress prediction into the same framework. Using logistic regression models trained on data from a Slovenian population-based study, we achieved AUROC values of 0.84, 0.80, and 0.82 for predicting depression, anxiety, and stress, respectively, alongside 0.82 for SI. Key predictors included behavioral disengagement, self-blame, and denial, while protective factors such as satisfaction with relationships were strongly associated with reduced risk. This model, requiring less than 5 min to administer, allows for early identification of individuals experiencing mental distress, facilitating timely intervention. Given the substantial burden of mental health challenges and their association with suicidality, integrating ML-based tools into public health strategies could significantly enhance early detection and intervention efforts, ultimately reducing preventable deaths.

External IDs:dblp:conf/aime/MaticSPP25