The process of identifying and selecting the most relevant characteristics or attributes from raw data to improve machine learning model performance and reduce computational complexity.