Catching Mental Health Crises Before They Escalate

Detecting behavioral health crises in 911 narratives before they escalate to jail or the ER

The 911 Mental Health Crisis Response project builds NLP and AI models that classify behavioral health content in messy, unstructured first-responder narratives, subject to real ethical constraints around transparency and explainability. Co-directed through the Modular Agile Deployment (MAD) Lab at Kennesaw State University with graduate students and faculty from Computer Science and Social Work, the project has produced peer-reviewed publications in Smart Health and presentations at IEEE BigData and the IEEE/ACM Conference on Connected Health.

Beyond the models themselves, the project required building new secure-enclave computing infrastructure at KSU for highly sensitive identified criminal and health data — infrastructure that now supports the university’s broader R2 research designation. PhD student Martin Brown successfully defended his dissertation on the project in 2024, achieving .99 classification accuracy through novel ensemble and active-learning techniques; fellow researcher Adnan Azmee won the CCSE Outstanding Researcher Award the same year. The project is funded through Partnership for Inclusive Innovation and Fitzgerald Foundation grants, has been presented to the Department of Homeland Security and the National Academies, and underlies a current National Science Foundation Smart Connected Communities proposal.