Misleading modelling: overfitting, cross-validation, and the bias-variance trade-off