Table 5. Results of the proposed methods (DBI, BRI & BRIX) on the USPS dataset with different reduction ratios.
Ratio | Training Speedup | Testing Speedup | Accuracy | Number of SVs | Jaccard similarity | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Whole dataset | 1.00 | 1.00 | 0.922 | 790 | 1.00 | ||||||||||
DBI | BRI | BRIX | DBI | BRI | BRIX | DBI | BRI | BRIX | DBI | BRI | BRIX | DBI | BRI | BRIX | |
0.01 | 88.81 | 90.70 | 98.19 | 6.24 | 8.81 | 9.44 | 0.488 | 0.305 | 0.786 | 82 | 58 | 52 | 0.08 | 0.07 | 0.02 |
0.02 | 66.88 | 64.54 | 72.90 | 4.37 | 5.06 | 6.77 | 0.548 | 0.701 | 0.896 | 139 | 113 | 80 | 0.12 | 0.09 | 0.04 |
0.03 | 58.51 | 62.66 | 63.79 | 3.92 | 4.22 | 5.85 | 0.564 | 0.802 | 0.909 | 178 | 161 | 102 | 0.15 | 0.10 | 0.04 |
0.04 | 54.48 | 50.28 | 62.17 | 3.29 | 3.41 | 5.78 | 0.622 | 0.871 | 0.915 | 217 | 199 | 109 | 0.18 | 0.12 | 0.05 |
0.05 | 39.06 | 45.14 | 51.77 | 2.52 | 3.20 | 5.12 | 0.684 | 0.891 | 0.917 | 254 | 227 | 123 | 0.20 | 0.12 | 0.05 |
0.1 | 24.22 | 27.71 | 33.35 | 1.70 | 2.38 | 4.84 | 0.850 | 0.917 | 0.920 | 405 | 306 | 142 | 0.25 | 0.15 | 0.07 |
0.2 | 13.03 | 14.96 | 18.85 | 1.79 | 2.38 | 4.38 | 0.911 | 0.921 | 0.921 | 412 | 309 | 159 | 0.29 | 0.19 | 0.09 |
0.3 | 9.08 | 9.15 | 12.87 | 1.76 | 2.19 | 4.15 | 0.917 | 0.921 | 0.921 | 415 | 326 | 179 | 0.33 | 0.23 | 0.12 |
0.4 | 6.73 | 6.57 | 8.20 | 1.85 | 2.16 | 3.20 | 0.919 | 0.921 | 0.921 | 421 | 345 | 225 | 0.34 | 0.26 | 0.17 |
0.5 | 4.99 | 5.14 | 6.57 | 1.80 | 2.17 | 3.00 | 0.920 | 0.922 | 0.921 | 418 | 362 | 248 | 0.35 | 0.29 | 0.20 |
0.6 | 4.21 | 4.38 | 5.05 | 1.81 | 2.09 | 2.77 | 0.920 | 0.922 | 0.922 | 406 | 374 | 272 | 0.36 | 0.32 | 0.23 |
0.7 | 3.44 | 3.31 | 3.75 | 1.83 | 1.90 | 2.39 | 0.921 | 0.922 | 0.921 | 403 | 388 | 307 | 0.38 | 0.35 | 0.27 |
0.8 | 2.72 | 3.01 | 2.84 | 1.78 | 1.87 | 2.09 | 0.921 | 0.922 | 0.922 | 407 | 397 | 345 | 0.39 | 0.37 | 0.32 |
0.9 | 2.34 | 2.52 | 2.37 | 1.74 | 1.78 | 1.97 | 0.922 | 0.922 | 0.922 | 411 | 405 | 383 | 0.41 | 0.40 | 0.38 |
Each row compares performance metrics of the proposed methods for a different value (0.01 to 0.05 and 0.1 to 0.9) of the reduction ratio parameter r (the user-defined desired proportion of the training set to retain). The first row represents the metrics measured after training on the whole dataset.