OBJECTIVES:  To develop and validate a predictive model for 12-month hypoglycemia risk in adults with T2D using an ensemble-based, hypothesis-free, predictive-modeling approach. METHODS:  Data were collected retrospectively for the period January 2008-December 2013 from the Truven Health MarketScan® Commercial, Medicare Supplemental, and Laboratory databases. Eligible participants were ≥18 years with T2D, and had a first pharmacy claim for an antidiabetes medication (index date) and continuous 18-month enrollment (6 months pre- and 12 months post-index). Baseline patient characteristics were collected during the pre-index period. Hypoglycemia events in the post-index period were identified via ICD-9 diagnosis codes or a blood glucose measure ≤70 mg/dL (Level 1 hypoglycemia according to 2016 ADA/EASD guidelines). Bayesian model averaging was via Markov Chain Monte Carlo sampling across the posterior distribution of model probabilities. This results in an ensemble of predictive models to examine the association of patients’ baseline characteristics and hypoglycemia risk. RESULTS: An ensemble of 128 predictive models for 12-month hypoglycemia risk was sampled from the posterior distribution of model probabilities based on data from 558,963 eligible participants. Overall, 5,823 patients (1.0%) had ≥1 hypoglycemia event captured during 12-months’ follow-up. The most probable predictors of hypoglycemia included prior hypoglycemia, older age (≥75 years), insulin use, sulfonylurea use, mood disorders, and high baseline healthcare utilization and costs. C-statistics were used to select the most accurate logistic regression model in a validation dataset (C-statistic=0.73). In a held-out sample that was used to test the likely accuracy of the model, participants were classified into two annual hypoglycemia-risk groups: ≤5.0% or >5.0%. The mean estimated hypoglycemia risk of the ≤5.0% group was 1.3% (95% CI 1.2-1.4%); the mean risk estimate of the >5.0% group was 14.0% (95% CI 12.5-15.6%). CONCLUSIONS:  Predictive modeling was successfully used to identify 12-month hypoglycemia risk and predictors of hypoglycemia in adults with T2D.