Gene Selection via a New Hybrid Ant Colony Optimization Algorithm for Cancer Classification in High-Dimensional Data

Table 6

Comparison of SVM, 1NN, MWIS-1NN, and MWIS-ACO-LS (LOOCV).

Datasets	Performance	SVM	LR-L1	Avg	Best	LR-Elasticnet	Avg	1NN	MWIS	MWIS-ACO	MWIS-ACO-LS	Best
11_Tumors	Accuracy (%)	85.63	88.22	93.1	86.38	88,22	74,14	67,24	94,90	96	99,14	99,42
	Genes	12533	—	—	—	—	12533	1308	463,00	460	166,9	101
	Time (min)	—	—	—	—	—	—	0,82	91,33	—	123,2	—
9_Tumors	Accuracy (%)	38.33	35.00	50.00	29.50	38,33	40,00	60,00	98,83	100	100,00	100,00
	Genes	5726	—	—	—	—	5726	263	90,10	83	51	40
	Time (min)	—	—	—	—	—	—	0,34	21,3	—	34,48	—
Brain_Tumor1	Accuracy (%)	88.89	85.67	88.89	85.44	88,89	85,56	80,00	96,56	100,00	99,22	100,00
	Genes	5920	—	—	—	—	5920	246	55,90	46	22,9	19
	Time (min)	—	—	—	—	—	—	0,22	29,14	—	45,81	—
Brain_Tumor2	Accuracy (%)	70.00	27.20	32.00	29.20	36,00	60,00	48,00	95,40	100,00	99,40	100,00
	Genes	10367	—	—	—	—	10367	110	22,40	18	11,1	11
	Time (min)	—	—	—	—	—	—	0,55	17,89	—	27,13	—
Leukemia1	Accuracy (%)	97.22	91.81	94.44	92.64	95,83	83,33	66,67	100,00	100,00	100,00	100,00
	Genes	5327	—	—	—	—	5327	297	63,00	56	9,4	5
	Time (min)	—	—	—	—	—	—	0,26	25,94	—	43,77	—
Leukemia2	Accuracy (%)	97.22	91.81	95,83	91.67	95.83	86.11	73.61	100.00	100.00	100.00	100.00
	Genes	11225	—	—	—	—	11225	203	45.80	42	13.9	11
	Time (min)	—	—	—	—	—	—	0,44	23,85	—	37.67	—
Lung_Cancer	Accuracy (%)	95.07	91.67	94.10	91.03	93.60	87.68	90.15	98.42	99.01	98.92	99.51
	Genes	12600	—	—	—	—	12600	602	180,00	183	34,8	36
	Time (min)	—	—	—	—	—	—	0,82	74,61	—	107,3	—
SRBCT	Accuracy (%)	100.00	97.59	100.00	97.10	98.79	85,54	91,57	100,00	100,00	100,00	100,00
	Genes	2308	—	—	—	—	2308	109	15,60	15	7,6	6
	Time (min)	—	—	—	—	—	—	0,14	23,48	—	38,24	—
Prostate_Tumor	Accuracy (%)	92.16	81.47	88.23	80.69	83.33	82,35	79,41	98,24	99,04	99,12	100,00
	Genes	10509	—	—	—	—	10509	193	47,30	43	20,3	21
	Time (min)	—	—	—	—	—	—	0,39	29,2	—	46,59	—
DLBCL	Accuracy (%)	97.40	91.95	97.40	91.43	94.80	84,42	87,01	100,00	100,00	100,00	100,00
	Genes	5469	—	—	—	—	5469	147	25,7	22	7,2	6
	Time (min)	—	—	—	—	—	—	0,17	27,18	—	43,63	—

Note: the best results are shown in bold. Remark: as the SVM, 1NN, and MWIS are of deterministic nature, the classification is calculated just in one run. Accuracy: the classification accuracy using LOOCV (leave-one-out-cross-validation). Genes: the number of genes used in the classification ofthe LR-L₁ and LR-Elasticnet methods. Best: the best result found in all ten runs. Avg: the average of the ten experiments. Time: the execution time in minutes. SVM: the support vector machine classifier using a linear kernel. LR-L₁: the logistic regression classifier with the lasso regularisation. LR-Elasticnet: the logistic regression classifier with the elastic net regularisation. 1NN: the 1-nearest neighbor classifier. MWIS: the maximum weight independent set for gene selection. MWIS-ACO: our method of selection combining MWIS and ACO without using LS. MWIS-ACO-LS: our improved method of selection combining MWIS and ACO and the local search algorithm (LS).