A hybrid machine learning approach that combines a small set of labeled training data with a larger set of unlabeled data. Useful for AML where only a few confirmed suspicious cases exist.