Combining multiple types of features (e.g., autoencoder outputs and statistical measures) into a single representation for improved model performance.