|
Validation of two subsidiary improvements and comparative experiments between GA and RBFN including above-mentioned improvements. The comparison between kNN and ADSGA are shown in the columns headed kNN and the second columns from right headed ADSGA. The results acquired with the conventional kNN method and our proposed ADSGA are shown in the columns headed kNN and ADSGA. The improved results obtained with the fitness transformation are demonstrated in the 2 columns headed ADSGA and ADSGA+Log. The validation of main improvement that is the application of RBFN is presented in the leftmost GA column and the rightmost ADSGA+Log column. The proposed method yielded equivalent or more accurate results compared to the parameters obtained with GA at half the calculation time and a 50% increase in the optimization success rate. |
|||||||
| GA |
RBFN |
||||||
|
|
|||||||
| kNN |
ADSGA |
ADSGA Log |
|||||
| k = 2 |
k = 4 |
k = 8 |
k = 16 |
||||
|
|
|||||||
| Convergence rate (%) |
60 |
90 |
88 |
92 |
90 |
86 |
90 |
| Processing time (min) |
273 |
163 |
188 |
170 |
157 |
167 |
130 |
| Test error (%) |
10.3 ± 2.5 |
12.9 ± 3.9 |
10.4 ± 2.0 |
11.0 ± 2.3 |
11.8 ± 3.4 |
9.3 ± 2.2 |
10.8 ± 2.1 |
Matsubara et al. BMC Bioinformatics 2006 7:230 doi:10.1186/1471-2105-7-230 |
|||||||