TLiSVM (Triple Linear SVM Weight) for dimensionality reduction
Wipawan Buathong · 2016
This article proposed `TLiSVM' or `3LiSVM' (Triple Linear SVM Weight) as an alternative technique for dimensionality reduction with a Support Vector Machine (SVM) classifier on a two-class dataset. The efficiency of TLiSVM was compared with two chosen techniques, including Linear SVM Weight (LiSVM) and Double Linear SVM Weight (DLiSVM). Three datasets, including DLBCL, Duke Breast-Cancer and Leukemia, were used for the experiment. The proposed technique was discovered to be more efficient than using either LiSVM or DLiSVM for dimensionality reduction. The accuracy rate could reach 100 percent in all experimental datasets with the same consistency of dimensionality reduction. While the dimensional data of DLBCL could be downsized from 7,070 to 16 features, the dimensional data of Duke Breast-Cancer and Leukemia could be downsized from 7,129 to 11 features.