BACKGROUND: Inflammatory bowel disease (IBD) is associated with increased health care utilization. Forecasting of high resource utilizers could improve resource allocation. In this study, we aimed to develop machine learning models (1) to cluster patients according to clinical utilization patterns and (2) to predict longitudinal utilization patterns based on readily available baseline clinical characteristics. METHODS: We conducted a retrospective study of adults with IBD at 2 academic centers between 2015 and 2021. Outcomes included different clinical encounters, new prescriptions of corticosteroids, and initiation of biologic therapy. Machine learning models were developed to characterize health care utilization. Poisson regression compared frequencies of clinical encounters. RESULTS: A total of 1174 IBD patients were followed for more than 5673 12-month observational windows. The clustering method separated patients according to low, medium, and high resource utilizers. In Poisson regression models, compared with low resource utilizers, moderate and high resource utilizers had significantly higher rates of each encounter type. Comparing moderate and high resource utilizers, the latter had greater utilization of each encounter type, except for telephone encounters and biologic therapy initiation. Machine learning models predicted longitudinal health care utilization with 81% to 85% accuracy (area under the receiver operating characteristic curve 0.84-0.90); these were superior to ordinal regression and random choice methods. CONCLUSION: Machine learning models were able to cluster individuals according to relative health care resource utilization and to accurately predict longitudinal resource utilization using baseline clinical factors. Integration of such models into the electronic medical records could provide a powerful semiautomated tool to guide patient risk assessment, targeted care coordination, and more efficient resource allocation.