There are complex and interconnected factors associated with suicide risk in serving and ex-serving military members. However, just as in the broader community, the causes of suicidal thoughts and behaviours are still not fully understood.

Consequently, there is considerable motivation to utilise techniques that may improve the identification of those at increased risk of suicide, including through machine learning (ML) techniques. Machine Learning, a type of artificial intelligence, creates systems and algorithms that can learn from and make predictions or decisions based on the available data, and the techniques are particularly useful in analysing large, unprocessed datasets. More traditional classification methods, such as logistical regression, require data analysis to be guided by the researcher through the application of a hypothesis. The application of ML to suicide research has been viewed as an opportunity to discover new areas for future suicide research by identifying predictors and their combinations from large, multi-domain datasets. The analysis is driven by the data, rather than a particular theory or hypothesis.

The primary aim of this exploratory study was to use ML techniques to identify which variables (survey items) best distinguished those who reported suicidal ideation only versus those who reported action (planning and or attempts), with or without ideation at a given point in time (i.e., estimated independently for the 2010 and the 2015 datasets) (Analysis 1 – cross-sectional data) and across time (Analysis 2 - longitudinal data).

A secondary aim of the study was to assess how well the ML techniques performed in building predictive models from the available data, compared to more traditional statistical methods, such as logistic regressions. Unlike ML which learns from the available data, logistic regression is a statistical technique which requires testing of a hypothesis in relation to which variables will predict the outcomes.