Page 81 - 《软件学报》2020年第12期

张鑫 等:自然进化策略的特征选择算法研究 3747

对比分类准确率,通过表 5~表 7 不难看出:当分类器和数据集划分方法相同时,对比其他所有算法,MCC-
NES 在 Glass,Heart,Cleveland,Dermatology,Spambase,Musk2,LSVT,SRBCT,Arcene,RNA-Seq,Dorothea 等 11 个数
据集上,其分类准确率都是最高的;而在 Wine 数据集的 Rbf-SVM 分类器上,分类准确率仅低于 HGAFS 算法
0.6%;在 Vehicle 数据集的 Rbf-SVM 分类器上,低于 HGAFS 算法 3%;在 Segmentatin 数据集的 3-NN 分类器上,
低于 FAFOA 算法 2.6%;在 Ionosphere 数据集的 Rbf-SVM 分类器上,低于 ISEDBFO 算法 0.9%;在 Sonar 数据集
的 Rbf-SVM 分类器上,低于 ACBFO 算法 12%.其中:在 Sonar 数据集中,当分类器为 1NN 时,MCC-NES 分类准
确率比 SBS 高出 35%.
Table 5 Classification accuracy and dimension reduction of MCC-NESand compared methods (1)
表 5 MCC-NES 及其对比算法的分类准确率和维度缩减率(1)
Glass CA (%) DR (%) Classifier
MCC-NES 85.23(70%~30%) 67.11 1-NN
FSFOA 71.88(70%~30%) 40.0 1-NN
SFS 72.24(70%~30%) 26.66 1-NN
SFFS 71.77(70%~30%) 37.77 1-NN
IFS-CoCo 55.44(10-fold) 41.11 1-NN
MCC-NES 70.61(2-fold) 55.56 RBF-SVM
FSFOA 68.22(2-fold) 60.0 RBF-SVM
HGAFS 65.51(2-fold) 44.44 RBF-SVM
MCC-NES 80.39(10-fold) 44.44 CART
FSFOA 75.7(10-fold) 50.0 CART
FS-NEIR 68.53(10-fold) 22.22 CART
Heart CA (%) DR (%) Classifier
MCC-NES 85.56(10-fold) 61.53 3-NN
FSFOA 85.18(10-fold) 35.71 3-NN
NSM 84.0(10-fold) 69.23 3-NN
MCC-NES 85.18(2-fold) 53.84 RBF-SVM
FSFOA 84.07(2-fold) 50.0 RBF-SVM
HGAFS 82.59(2-fold) 76.92 RBF-SVM
MCC-NES 85.18(10-fold) 76.92 CART
FSFOA 85.15(10-fold) 48.07 CART
FS-NEIR 79.86(10-fold) 46.15 CART
Cleveland CA (%) DR (%) Classifier
MCC-NES 67.11(70%~30%) 74.61 1-NN
FSFOA 55.55(70%~30%) 71.42 1-NN
SVM-FuzCoc 61.01(70%~30%) 46.1 1-NN
SFS 51.79(70%~30%) 47.7 1-NN
SBS 54.80(70%~30%) 38.5 1-NN
SFFS 49.50(70%~30%) 53.8 1-NN
Wine CA (%) DR (%) Classifier
MCC-NES 99.53(70%~30%) 69.23 5-NN
FSFOA 99.20(70%~30%) 30.76 5-NN
PSO(4-2) 95.26(10-fold) 51.6 5-NN
FW-NSGA-II 98.98(10-fold) 53.84 5-NN
MCC-NES 99.44(70%~30%) 66.15 1-NN
FSFOA 98.07(70%~30%) 50.0 1-NN
SVM-FuzCoc 97.12(70%~30%) 53.84 1-NN
SFS 97.69(70%~30%) 35.38 1-NN
SBS 94.77(70%~30%) 46.15 1-NN
SFFS 96.56(70%~30%) 36.92 1-NN
MCC-NES 97.75(2-fold) 53.85 RBF-SVM
FSFOA 96.06(2-fold) 37.17 RBF-SVM
HGAFS 98.31(2-fold) 53.85 RBF-SVM
MCC-NES 99.26(70%~30%) 61.53 CART
FSFOA 96.0(70%~30%) 57.14 CART
UFSACO 95.08(70%~30%) 61.53 CART
MCC-NES 97.70(10-fold) 61.53 CART
FSFOA 96.06(10-fold) 21.42 CART
FS-NEIR 95.04(10-fold) 61.53 CART
FW-NSGA-II 98.62(10-fold) 57.69 CART