Meta-analyses indicate that efforts to predict self-harming behaviors have resulted in near-chance accuracy. Prediction of these behaviors may benefit from machine learning methods, which can better account for the complexity of self-harming behaviors. My work in this area uses machine learning to improve the longitudinal prediction of nonsuicidal self-injury (NSSI), suicide, and eating disorders across both short-term periods of high risk and long-term studies of illness course and outcome.