Hyperparameter tuning is the process of selecting settings that control model training or architecture before or during experimentation.
Examples include learning rate, batch size, regularization strength, tree depth, and number of layers. Tuning must be separated from final test evaluation to avoid overstating performance.