Suicide is a serious public health problem worldwide, and earlier identification of risk could help get people support sooner. This research study explored whether machine learning can help identify patterns linked to suicidality.
The problem
Earlier identification of suicide risk could help connect people with support and reduce deaths. Researchers are exploring whether data-driven methods can complement clinical judgement.
What the research did
We compared four machine learning techniques, Random Forest, XGBoost, Support Vector Machines and LightGBM, on data about suicidality in different groups. A Select-K-Best feature scoring step chose the 12 most informative characteristics to improve performance.
What we found
On the test data, XGBoost achieved the best accuracy, at 89%.
Who benefits
- Mental health researchers studying risk factors
- Clinicians, potentially, as one input alongside professional assessment
- Public health planners
Limits and what comes next
This is an early research study, not a clinical tool. Suicide risk is complex and individual; no algorithm can replace a conversation with a trained professional. Tools like this raise important questions about privacy, consent and the harm of errors, which must be addressed before any real-world use.
If you or someone you know is struggling, please reach out for support. In Australia, Lifeline is available 24/7 on 13 11 14. Elsewhere, contact your local emergency number or a crisis helpline.
The team
This research was led by Md Naimur Rahaman and colleagues at Daffodil International University, with me.
Frequently asked questions
Can AI predict suicide?
No algorithm can reliably predict an individual’s actions. Research like this explores patterns that might help professionals, but it is not a diagnostic tool.
What did the study find?
Of four methods tested, XGBoost was most accurate on the study data, at 89%.
Where can people get help?
In Australia, Lifeline is available 24/7 on 13 11 14. Elsewhere, contact local emergency services or a crisis line.