It involves training a model on a dataset without any labeled data. The model learns to identify patterns and relationships in the data without any prior knowledge of the output variables. It is often used in clustering, anomaly detection, and pattern recognition problems.