My work fundamentally advances the predictability of risky health behaviors by using state-of-the-art approaches to ubiquitous computing and machine learning in the areas of health monitoring and support systems, particularly for at-risk populations. I have successfully developed a mobile AI systems that predicts 30-day hospital readmissions for post-surgical cancer patients and expanded the system to estimate the severity of symptoms during chemotherapy, high-risk alcohol consumption, and marijuana intoxication predictions in young adults (Best paper in Cancer Informatics, published in ACM and Forbes news). My long-term research goal is to establish new knowledge of a personalized intervention system to empower at-risk populations leveraging mobile sensing, machine learning and Human-centered AI (HAI) design strategies for human behavior change in health and safety.