Name of the Method | Key Idea | Performance Analysis |
---|---|---|
Euclidian distance and Pearson correlation feature selection [66] | • Euclidian distance and Pearson correlation coefficient for feature selection • SVM classifier with different kernel |
Distance-based method outperforms for SVM with a linear kernel. |
SNR [65], t-test [121] | • k-means for attribute clustering • Signal to noise ratio and t-statistics for feature selection • SVM, kNN, PNN, FNN are used for classification. . |
The performance of SVM classifier gives a better result for the 5 features using k-means-SNR and k-means-t-test approach. |
Filter approach [63] | • IG, RA, TA, and PCA for feature selection. • SVM, kNN, DT, NB, and NN for classification. |
The best classification accuracy is achieved by using a subset of 250 features chosen by IG based method for SVM classifier. |
mRMR [72] | • mRMR for feature selection • NB, SVM, LDA for classification |
The computational cost of mRMR is low and the classification accuracy is high in comparison to maximum dependency and maximum relevance for all datasets. |
Best Incremental Ranks Subset(BIRS) [74] | • BIRS (wrapper), nonlinear correlation measure based entropy and IG feature selection methods. • NB,IB,C4 for classification |
The computational complexity of BIRS is better in comparison to CFS, FCBF. |
Kernel panelized SVM(KP-SVM) [80] | • KP-SVM for feature selection • SVM for classification |
The advantage of KP-SVM in terms of computational effort is that it automatically obtains an optimal feature subset, avoiding a validation step to determine how many ranked features will be used for classification. |