Semi-supervised learning is a machine learning technique that trains a predictive model using supervised learning, a small set of labeled data, and a large set of unlabeled data.