The use of machine learning in suicide research and prevention is often discussed in terms of its ‘great promise’. As yet, it is unclear to what extent we can use machine learning to deliver on those promises and whether adequate consideration has been given to ethical, transparency, and clinical implementation issues. This talk will cover a selection of these key issues, drawing on a recent review of the machine learning and suicide research literature.
These considerations include how to meet the potentially increased need for interventions arising from improving identification of individuals at risk of making an attempt and who will be responsible for this, fairness and bias in the application of machine learning, the need for sufficient and high-quality data, and issues around interpretability, transparency, and reproducibility of machine learning models in suicide research. I will discuss the implications of these issues for research and practice in suicide prevention, as well as highlighting applications of machine learning in suicide prevention, such as intervention selection and clinician training, that may be more likely to deliver on its promises than risk prediction.