Feature learning is a technique in machine learning where the system automatically discovers the representations needed for feature detection or classification from raw data. It enables models to improve performance by learning to transform input data into a more useful representation without explicit manual feature engineering.