An improved clear cell renal cell carcinoma stage prediction model based on gene sets

BMC Bioinformatics

Table 3 The performance of machine learning-based models developed using different sets of selected features, which include whole gene sets without feature selection, RCSP-set-Weka-Hall, FCBF-set, and FJL-set

Features	Algorithms	Methods	Performance Measures
Features	Algorithms	Methods	Sensitivity	Specificity	Accuracy(%)	MCC	AUC
Whole gene set (20,530 genes)	SVM	10-fold	0.182	0.943	63.25	0.198	0.709
	SVM	Testing	0.020	1.000	60.66	0.111	0.806
	LR	10-fold	0.590	0.777	69.91	0.370	0.683
	LR	Testing	0.673	0.863	78.69	0.551	0.768
RCSP-set-Weka-Hall (38 genes)	SVM	10-fold	0.696	0.697	70.35	0.386	0.769
RCSP-set-Weka-Hall (38 genes)	SVM	Testing	0.735	0.808	77.87	0.541	0.844
FCBF set (101 genes)	SVM	10-fold	0.727	0.758	74.23	0.475	0.793
	SVM	Testing	0.776	0.740	75.41	0.506	0.826
	LR	10-fold	0.678	0.742	71.57	0.415	0.768
	LR	Testing	0.612	0.808	72.95	0.429	0.789
FJL set (23 genes)	Discretization +SVM	10-fold	0.680	0.868	79.27	0.562	0.852
	Discretization +SVM	Testing	0.714	0.877	81.15	0.603	0.860
	Discretization + LR	10-fold	0.750	0.805	78.45	0.556	0.855
	Discretization + LR	Testing	0.756	0.767	77.87	0.554	0.860
	Discretization +SVM	100 random test sets	0.710	0.788	75.64	0.496	0.831
	Discretization + LR	100 random test sets	0.647	0.876	78.32	0.542	0.842

ISSN: 1471-2105