Table 2 Error rate comparison results on classification benchmark datasets (mean ± standard deviation). The best value in each row is shown in bold, and the second-best value is underlined.
From: Test-time local training of neural network for tabular data
Dataset | No. instances \(\times\) no. features (no. classes) | Local learning | Global learning | Proposed | |||
|---|---|---|---|---|---|---|---|
k-NN | LWSR | SR | RF | NN\(_\text {global}\) | NN\(_\text {TTLT}\) | ||
Drybean | 13,611 \(\times\) 16 (7) | 0.0749 ± 0.0025 | 0.0704 ± 0.0023 | 0.0771 ± 0.0029 | 0.0758 ± 0.0031 | 0.0711 ± 0.0035 | 0.0683 ± 0.0029 |
Letter | 20,000 \(\times\) 16 (26) | 0.0473 ± 0.0032 | 0.0369 ± 0.0019 | 0.2255 ± 0.0044 | 0.0341 ± 0.0022 | 0.0253 ± 0.0026 | 0.0189 ± 0.0018** |
Magic | 19,020 \(\times\) 10 (2) | 0.1570 ± 0.0051 | 0.1319 ± 0.0049 | 0.2081 ± 0.0066 | 0.1200 ± 0.0059 | 0.1228 ± 0.0053 | 0.1247 ± 0.0042 |
Optdigits | 5620 \(\times\) 64 (10) | 0.0190 ± 0.0028 | 0.0927 ± 0.0061 | 0.0414 ± 0.0042 | 0.0180 ± 0.0029 | 0.0230 ± 0.0026 | 0.0147 ± 0.0031* |
Pageblocks | 5473 \(\times\) 10 (5) | 0.0335 ± 0.0049 | 0.0300 ± 0.0053 | 0.0384 ± 0.0052 | 0.0271 ± 0.0040 | 0.0301 ± 0.0054 | 0.0260 ± 0.0045* |
Penbased | 10,992 \(\times\) 20 (10) | 0.0056 ± 0.0011 | 0.0057 ± 0.0011 | 0.0454 ± 0.0039 | 0.0078 ± 0.0009 | 0.0057 ± 0.0012 | 0.0041 ± 0.0008** |
Satimage | 6435 \(\times\) 36 (6) | 0.0942 ± 0.0063 | 0.0923 ± 0.0067 | 0.1437 ± 0.0042 | 0.0869 ± 0.0076 | 0.0932 ± 0.0073 | 0.0788 ± 0.0070* |
Texture | 5500 \(\times\) 40 (11) | 0.0111 ± 0.0039 | 0.0021 ± 0.0014 | 0.0027 ± 0.0020 | 0.0209 ± 0.0035 | 0.0020 ± 0.0017 | 0.0012 ± 0.0007 |
Waveform | 5000 \(\times\) 21 (3) | 0.1480 ± 0.0094 | 0.1611 ± 0.0114 | 0.1322 ± 0.0093 | 0.1503 ± 0.0089 | 0.1322 ± 0.0081 | 0.1331 ± 0.0097 |
Average Rank | – | 4.22 ± 1.03 | 3.83 ± 1.29 | 5.17 ± 1.45 | 3.44 ± 1.71 | 2.89 ± 0.97 | 1.44 ± 0.83 |
