Table 2.

Most important five features for each of the eight feature preselection methods.

Feature pre-selection method	Rank of feature importance
	1	2	3	4	5
Stepwise logistic	Tumor shape: irregular or regular	Tumor location: convexity	Tumor location: falx	orig.shape.Elongation	fd_vs_re: first diagnose or relapse
Lasso	Tumor shape: irregular or regular	Tumor location: skull base	orig.shape.Elongation	fd_vs_re: first diagnose or relapse	Tumor location: posterior fossa
Ridge	Tumor shape: irregular or regular	Tumor location: convexity	Tumor location: falx	Tumor location: skull base	orig,shape,Elongation
GBM	Tumor shape: irregular or regular	Tumor location: skull base	Tumor location: convexity	orig,glszm.SizeZone NonUniformity	orig,shape,Elongation
Random forest	Tumor location: convexity	Tumor location: skull base	orig.glszm.SizeZone NonUniformity	Shape: irregular or regular	fd_vs_re: first diagnose or relapse
Bagged trees	Tumor location: skull base	Tumor location: convexity	orig.shape.Sphericity	Shape: irregular or regular	orig.shape.Elongation
LDA	Tumor shape: irregular or regular	Tumor location: convexity	Tumor location: skull base	KPI	orig.glszm.Small AreaEmphasis
Naive Bayes	Tumor shape: irregular or regular	Tumor location: convexity	Tumor location: skull base	KPI	orig.glszm.Small AreaEmphasis