Table 1:

Overview on traditional gene selection approaches and their classification. Filter approaches have lowest complexity, wrapper and embedded approaches apply machine learning strategies, hybrid and ensemble approaches combine multiple approaches.

Category	Functionality	Characteristics	Approaches
filter	uses only	+ independent of classifier	Information Gain (IG) [8]
	intrinsic data	+ low complexity	ReliefF [9]
	characteristics	+ good generalization	and [10], [11], [12], [13], [14], [15]
wrapper	uses learning	+ detects gene dependencies	Genetic Algorithms (GA) [16]
	algorithm to	/ interacts with classifier	Successive Feature
	evaluate genes	− high complexity	Selection (SFS) [17]
		− risk of overfitting
embedded	gene selection	+ detects gene dependencies	Support Vector Machine
	is embedded into	/ interacts with classifier	with Recursive Feature
	learning algorithm		Elimination(SVM-RFE) [18] and [19], [20]
hybrid	applies multiple	/ intermediate complexity	SVM-RFE + mRMR Filter [21]
	approaches	− risk of slight overfitting	and [22], [23]
	sequentially
ensemble	uses aggregate	+ good for small	Ensemble Gene Selection
	of a group of	sample domains	by Grouping (EGSG) [24]
	gene rankings	+ less prone to overfitting	and [25], [26]
		− computationally expensive
		− difficult to interpret