Accuracy result of feature selection (achieved on miRBase16). Here, we treat the feature selection as a matrix dimension reduction issue and considered three different methods: LLE, LSA and Isomap. The horizontal axis is the reduced dimension, and the vertical axis is the clustering accuracy. The "Origin" line stands for the performance of clustering result with all n-gram features, while others are results of accuracy after feature selection with different methods.
Wan et al. BMC Genomics 2012 13(Suppl 8):S15 doi:10.1186/1471-2164-13-S8-S15