Predicting Learning in a Multi-component Serious Game