Table 8 Regression performance of the ml-MRMR Methods for the German database when the minimum occurrence threshold Nminis sets 3 and 5. Results are shown both when the RCR algorithm is used and not used. In the underlined bold case, improvement is significant compared to the baseline system. Result with Gemaps feature set was: 10.14
Dim | Baseline | Nmin=5 | Nmin=5 and RCR | Nmin=3 | Nmin=3 and RCR | RCR |
|---|---|---|---|---|---|---|
(Ger+Tr) | (Ger+Tr) | (Ger+Tr) | (Ger+Tr) | |||
10 | 9.90 | 9.97 | 9.99 | 10.37 | 12.39 | 10.02 |
15 | 9.81 | 10.21 | 9.43 | 10.13 | 12.12 | 10.08 |
20 | 9.86 | 10.32 | 9.52 | 9.84 | 10.68 | 9.74 |
40 | 10.25 | 10.35 | 10.45 | 9.73 | 11.36 | 10.22 |
80 | 10.69 | 9.93 | 9.88 | 9.42 | 10.93 | 10.06 |
100 | 10.48 | 9.93 | 9.74 | 9.50 | 10.54 | 10.17 |
200 | 10.12 | 10.00 | 10.38 | 9.69 | 10.28 | 10.44 |
400 | 10.14 | 9.79 | 10.21 | 9.58 | 10.29 | 10.13 |
800 | 10.08 | 9.86 | 10.11 | 9.91 | 10.11 | 9.89 |
1000 | 10.02 | 9.85 | 10.14 | 9.79 | 10.16 | 9.98 |