Table 5.

A category of different feature selection methods, their advantages and limitations.

Methods		Advantages	Limitations	Example
Wrapper methods	Deterministic	-Simple -Dependence to feature -Interplay with classifier -Slower than Randomize	-High risk to over-fitting -More entrapment to local optimum than Randomize -Classifier dependent selection	-Sequential forward selection (SFS) -Sequential backward elimination (SBE) -Beam search
	Randomize	-Dependence to feature -Less entrapment to local optimum -Interplay with classifier	-Classifier dependent selection -More risk of over-fitting than deterministic	-Simulated Annealing -Randomized hill climbing -Genetic algorithms -Estimation of distribution algorithms
Filter methods	Univariate	-Quick -Gradable -No dependence to the classifier	-Relinquish dependence to feature -Relinquish interplay with the classifier	-Information Gain (IG) -x2 − CHI -t-test
	Multivariate	-Dependence to feature -No dependence to the classifier -Better time complexity than wrapper	-Slower than univariate methods -Less gradable than univariate methods -Relinquish interplay with the classifier	-Correlation-based feature selection (CFS) -Markov blanket filter (MBF) -Fast correlation-based feature selection (FCBF)
Embedded methods		-Dependence to feature -Interplay with classifier -Better time complexity than wrapper	-Classifier dependent selection	-Decision trees -Weighted naive Bayes -Feature selection using the weight vector of SVM