This study will determine whether suicidal civilians, active duty service members, and veterans differ or are essentially the same using machine learning algorithms based on approximately fifty of the most theoretically relevant MSRC Common Data Elements. Additional analysis will compare results for suicidal ideators and attempters.

Although there is likely some overlap among the processes associated with suicidality across active duty individuals, veterans, and civilians, there may be important differences as well. As the military suicide rate increased beyond the civilian rate around 2008, the differences between these groups became a larger topic of interest. Several theories have been proposed as reasons for the increased rates of suicidality in the military. Unfortunately, the majority of studies do not support these theories. One potential reason for these findings is that existing studies focus upon single factors or small sets of factors to distinguish between military and non-military samples. It is likely that large sets of factors will be necessary to accurately distinguish between military and non-military individuals with a history of suicidal thoughts or behaviors. Moreover, consistent with recent studies, this large set of factors must be combined in a complex, optimized fashion. Traditional statistical approaches are not well-suited for this task, but machine learning methods are well-suited because they were developed for these kinds of complex classification problems.

Drs. Franklin and Ribeiro are conducting a study to investigate whether machine learning algorithms can accurately distinguish between active duty servicemembers, veterans, and civilians with a history of suicidal thoughts or behaviors. Through utilizing the common data elements (CDE), a group of variables collected by the MSRC, and a machine learning method called Random Forests, Drs. Franklin and Ribeiro hope to separate these groups with at least ninety percent accuracy. This study will also use these methods to distinguish between suicide ideators and attempters within groups of active duty servicemembers, veterans, and civilians. It is expected that machine learning algorithms will be most accurate at distinguishing ideators from attempters in civilian populations compared to military populations. 

The results of this study could be beneficial in many ways. It can help clarify the differences between active duty servicemembers, veterans, and civilians who have suicidal thoughts or behaviors. It should provide further insight into the differences among suicide ideators and attempters for military and non-military groups. More broadly, this study will shed light on the degree to which processes associated with suicidality in civilians apply (or do not apply) to the processes associated with suicidality in military servicemembers.