Methods, systems, and apparatuses are described herein for protecting user accounts using machine learning models. A machine learning model may be trained to determine whether account activity indicates a risk to credit scores. Account data, associated with a first financial account, may processed to determine whether the first financial account is associated with at least one underage user. A transaction request, associated with the first financial account, may be received. A history of transactions conducted by the first financial account may be retrieved. The trained machine learning model may be provided, as input, the transaction request and the history of transactions. An indication of risk to a credit score associated with the at least one underage user may be received as output from the trained machine learning model. A limitation may be added to the first financial account based on the indication of risk.