FRL: An Integrative Feature Selection Algorithm Based on the Fisher Score, Recursive Feature Elimination, and Logistic Regression to Identify Potential Genomic Biomarkers

Table 4

Comparisons between FRL and other five feature selection algorithms.

Indicators	FRL		Ridge		Extra Trees		Random Forest		Lasso		Lasso-Logistic regression		Average of the other methods
	ACC	F1-score	ACC	F1-score	ACC	F1-score	ACC	F1-score	ACC	F1-score	ACC	F1-score	ACC	F1-score
GSE1420	100	100	75	70.83	62.5	60.45	75	70.83	62.5	62.5	75	75	70	67.92
GSE3325	100	100	83.33	85.19	66.67	70.83	100	100	83.33	75.76	83.33	85.19	83.33	83.39
GSE7696	100	100	96.15	94.27	100	100	96.15	94.27	96.15	94.27	100	100	97.69	96.56
GSE9750	100	94.9	96.15	83.73	100	84.7	96.15	89.52	65	65	100	100	91.46	84.59
GSE16088	95	100	85	100	85	100	90	100	83.33	75.76	100	100	88.67	95.15
GSE18520	100	100	100	90	100	100	100	95.18	89.47	84.5	94.74	95.18	96.84	92.97
GSE18842	100	92.86	85	85.57	95	78.57	94.74	89.24	53.57	53.57	89.29	89.24	82.38	81.37
GSE54129	100	100	85	78.11	100	100	100	100	85	78.11	100	100	94	91.24
GSE56315	100	100	100	100	100	100	100	100	100	100	100	100	100	100
GSE65194	100	100	100	100	90	89.05	98	97.81	66.67	66.67	98	97.81	90.53	90.27
Average	99.5	98.776	90.563	88.77	89.917	88.36	95.004	93.685	78.502	75.614	94.036	94.242	89.49	88.348