Skip to main content

Table 3 Comparison of maximum performance across processing components

From: Improving clustering performance using independent component analysis and unsupervised feature learning

Method Performance of applied processing (NMI, ACC)
L2 L2, ICA L2, RICA L2, RICA, ICA L2, SFT L2, SFT, ICA
COIL20 0.918, 0.857b 0.92, 0.856b 0.929, 0.894b 0.914, 0.885b 0.965, 0.93 b 0.946, 0.912a
COIL100 0.914, 0.774b 0.914, 0.784a 0.943, 0.813b 0.962, 0.897 a 0.932, 0.765b 0.954, 0.849a
CMU-PIE 0.941, 0.85b 0.986, 0.937 c 0.816, 0.716b 0.848, 0.721d 0.844, 0.759b 0.866, 0.774b
USPS 0.845, 0.828a 0.854, 0.81a 0.868, 0.926 a 0.85, 0.794e 0.852, 0.817b 0.853, 0.813b
MNIST 0.774, 0.787a 0.824, 0.882 b 0.79, 0.828b 0.787, 0.822b 0.79, 0.824b 0.794, 0.853a
REUTERS-10K 0.446, 0.656c 0.46, 0.714 c     
  1. Both NMI and ACC are presented in each cell, where the first value is the NMI. Italic font within a row indicates the maximum performance obtained for a dataset. The clustering algorithm providing the maximum performance for a given processing component and dataset is indicated by the following symbols: aGNMF; bSPC-SYM; cPCA; dSPC-RW; eICA-SYM