Mixed-Effects Machine Learning

The approach is intended to capture non-linear patterns, multivariate interactions, longitudinal structure, and clustered data that traditional models may miss. Model transparency will be supported through predictor importance plots and Shapley additive explanations. The core modelling approach is longitudinal mixed-effects machine learning using youth-provider relationship as the main predictor. Five studies will form the master training dataset, while MAGGIE will be held out as an independent test set. Random forest regression will be used for continuous outcomes and random forest classification for binary responder status.