Table 1.

Three representative machine learning methods with select pre-processing tips and tuning methods for complexity control.

Method	Pre-process	Complexity control	Reference
Support vector machine (SVM)	–Encode features as binary –Normalize to uniform distribution –Imputation for balancing data	–Recursive feature elimination for linear SVM –Soft margin width (C-parameter) –Kernel hyperparameters	(76, 160)
Bayesian networks	–Feature discretization –Variable selection to reduce graph search space –Imputation not necessary when using expectation maximization	–Constraints to a graph search space based on prior knowledge –Graph scoring functions that penalize complexity	(167–171)
Random forest	–No discretization or normalization necessary –Imputation required	–Number of features to sample at each node split (mtry) –Minimum number of samples in a terminal node	(172, 173)