Table 10: Soft computing methods performance analysis for microarray data

Name of the Method Key Idea Performance Analysis
A hybrid approach with GA wrapper [48] ▪ Ensemble feature selection using entropy-based feature ranking, t-statistics, and SVM-REF
▪ GA is applied to search an optimal or near optimal feature subset from the feature pool.
▪ SVM for classification
SVM-RFE shows better classification performance than other selection techniques for all datasets.
Redundant gene selection using PSO [122] ▪ Gene selection by RGS-PSO
▪ SVM,LG,C45,kNN,NB are used for classification
RGS-PSO and mRMR with 20genes are the best two methods, which have the top averaged BACC scores 0.818.
Rough set and SVM based [123] ▪ Rough set and MRMS for gene selection
▪ SVM for classification
The MRMS selects a set of miRNAs having a lowest B.632+ bootstrap error rate of the SVM classifier for all the data sets.
The better performance of the MRMS algorithm is achieved due to the use of rough sets.
ACO [52] ▪ ACO for feature selection
▪ In this paper, each gene is viewed as a node on the TSP problem. The nodes on the tour generated by the ant colony are the selected genes for cancer classification.
▪ BPNN for classification
.
ACO feature selection algorithm improves the performance of BPNN. Area under ROC curve (AUC) value after feature selection increased from 0.8531 to 0.911.
Rough set based [128] ▪ Rough set theory for feature selection by maximizing the relevance and significance of selected genes.
▪ K-NN and SVM for classification
The performance of proposed MRMS criterion is better than that of Max-Dependency and Max-Relevance criteria in most of the cases.
Out of total 28 cases, the MRMS criterion achieves significantly better results than Max-Dependency or Max-Relevance in 25 cases.