Risk of suicide continues to threaten both individual wellness and overall Air Force personnel readiness. The active-duty suicide rate increased 13% in 2018, while the suicide rate for veterans was 31 per 100,000 - 1.5 times greater than among nonveteran groups. Predictive analytics approaches using sensitive Electronic Health Records (EHRs) or social media data have shown promising accuracy results but must be augmented to improve sensitivity and specificity to acceptable levels for this critical challenge. Personnel systems offer military-relevant predictors of suicide risk (e.g., leave use patterns, deployment characteristics), but these have not yet been instantiated in a predictive model. The PARSERS effort will aid health professionals, human resources staff, chaplains, and others involved in facilitating Air Force personnel wellness by applying deep-learning architectures to personnel records to identify features of relevance for suicide prediction, incorporate these features into an explainable model for interpretability, and ensure that privacy-preserving methods are incorporated at all levels of analysis, given the sensitive nature of the data. The overall goal of the PARSERS effort is to provide technical and operational support that both provides the required analytics and emphasizes robust privacy preservation, enabling Airmen to perform at the highest level throughout their careers.
