Cross-validation techniques such as holdout, k-fold, stratified k-fold, and time series split provide methods to evaluate model performance on unseen data. K-fold reduces variance in estimates, stratified is suited for imbalanced classification, and time series split addresses temporal dependencies in data.