4. Average kNN accuracy on the Sina dataset.Figure optionsDownload full-size imageDownload as PowerPoint slide
For the Isolet6 dataset, the number of nearest neighbors is set to 3. From Fig. 2, we find that CA-074 Me CLE_LMNN performs best and achieves about 98% accuracy. LMNN achieves 97% accuracy and is better than other distance metrics except CLE_LMNN. The performance of ED is worse than CLE_LMNN and LMNN, but not worse than LMNN with unsupervised partition.