Constrained LVQ - Porownanie klasyfikatorów - Test 10CV
| Dataset | QPC/LVQ1 | PCA/LVQ1 | LVQ1 | SVM | kNN | MLP | |||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| vec | feat | cl | acc. | std. | #P | std. | #K | std. | acc. | std. | #P | std. | #K | std. | acc. | std. | #K | acc. | std. | #SV | std. | acc. | std. | #K | std. | acc. | std. | #N | |
| Appendicities | 106 | 7 | 2 | 86.1 | 8.5 | 1.0 | 0.0 | 2.5 | 1.2 | 82.2 | 7.4 | 1.0 | 0.0 | 3.1 | 1.4 | 86.6 | 6.8 | 2 | 86.7 | 10.8 | 31.4 | 2.9 | 84.0 | 6.1 | 4.8 | 1.7 | 87.6 | 9.3 | 0 |
| Australian | 690 | 14 | 2 | 86.1 | 4.1 | 1.0 | 0.0 | 2.0 | 0.0 | 86.1 | 4.1 | 1.0 | 0.0 | 2.0 | 0.0 | 85.6 | 4.6 | 2 | 84.9 | 1.6 | 206.1 | 4.6 | 85.1 | 2.9 | 8.2 | 1.9 | 87.0 | 5.6 | 1 |
| BitSymmetry 2 | 1000 | 20 | 2 | 87.3 | 4.2 | 1.2 | 0.4 | 3.5 | 1.0 | 78.1 | 11.4 | 1.8 | 0.7 | 5.3 | 1.7 | 75.7 | 4.3 | 12 | 98.0 | 1.4 | 296.5 | 7.8 | 89.3 | 2.6 | 5.8 | 6.2 | 96.7 | 2.4 | 2 |
| BitSymetry15 | 100 | 15 | 2 | 82.0 | 11.7 | 1.0 | 0.0 | 3.0 | 0.0 | 53.0 | 14.9 | 2.0 | 0.8 | 4.3 | 2.3 | 68.0 | 12.5 | 18 | 75.0 | 13.5 | 80.0 | 8.4 | 72.0 | 14.7 | 2.5 | 2.7 | 85.0 | 10.8 | 2 |
| BitSymetry20 | 100 | 20 | 2 | 81.0 | 7.0 | 1.0 | 0.0 | 3.0 | 0.0 | 58.0 | 9.8 | 1.9 | 0.8 | 4.2 | 1.3 | 59.0 | 15.8 | 8 | 70.0 | 14.2 | 84.7 | 6.9 | 68.0 | 11.3 | 2.9 | 2.8 | 76.0 | 15.1 | 2 |
| Breast Cancer W. | 683 | 9 | 2 | 96.2 | 1.8 | 1.0 | 0.0 | 2.0 | 0.0 | 96.2 | 2.1 | 1.0 | 0.0 | 2.0 | 0.0 | 96.5 | 2.5 | 2 | 96.6 | 1.7 | 51.1 | 3.1 | 97.1 | 1.4 | 5.6 | 2.1 | 96.6 | 1.4 | 0 |
| Czerniak (trs) | 250 | 14 | 4 | 85.7 | 6.2 | 1.0 | 0.0 | 4.0 | 0.0 | 70.8 | 9.0 | 1.7 | 0.6 | 4.3 | 0.6 | 76.8 | 7.2 | 4 | 85.2 | 5.7 | 240.3 | 23.0 | 86.0 | 7.4 | 1.0 | 0.0 | 96.0 | 2.7 | 4 |
| Glass | 214 | 9 | 6 | 60.4 | 10.3 | 1.1 | 0.3 | 4.5 | 0.9 | 58.9 | 7.3 | 1.4 | 0.6 | 4.0 | 0.5 | 66.3 | 10.5 | 7 | 64.9 | 6.2 | 283.6 | 17.6 | 68.8 | 9.2 | 1.4 | 0.8 | 69.2 | 10.4 | 7 |
| Heart | 270 | 13 | 2 | 78.9 | 8.5 | 1.0 | 0.0 | 2.0 | 0.0 | 80.7 | 9.0 | 1.0 | 0.0 | 2.0 | 0.0 | 82.2 | 8.6 | 2 | 81.5 | 9.1 | 101.5 | 7.0 | 78.5 | 7.6 | 8.5 | 1.2 | 82.6 | 6.8 | 0 |
| Ionosphere | 200 | 34 | 2 | 78.4 | 8.0 | 1.0 | 0.0 | 3.1 | 0.7 | 75.5 | 8.5 | 1.1 | 0.3 | 3.5 | 0.8 | 81.9 | 7.0 | 4 | 93.5 | 4.7 | 61.0 | 4.1 | 84.0 | 7.7 | 1.2 | 0.6 | 89.4 | 8.1 | 3 |
| Iris | 150 | 4 | 3 | 96.0 | 4.4 | 1.0 | 0.0 | 3.0 | 0.0 | 94.7 | 4.0 | 1.0 | 0.0 | 3.0 | 0.0 | 97.3 | 3.3 | 3 | 96.7 | 4.7 | 39.6 | 5.5 | 94.5 | 6.9 | 5.8 | 3.1 | 94.7 | 8.2 | 0 |
| L. Breast | 277 | 9 | 2 | 72.6 | 4.4 | 1.0 | 0.0 | 2.0 | 0.0 | 74.0 | 6.9 | 1.0 | 0.0 | 2.1 | 0.3 | 74.7 | 6.0 | 3 | 73.3 | 9.6 | 143.6 | 4.5 | 73.7 | 5.5 | 6.9 | 2.7 | 75.8 | 5.6 | 2 |
| Led500 | 500 | 7 | 10 | 58.5 | 8.0 | 1.1 | 0.3 | 9.6 | 0.5 | 45.5 | 8.6 | 1.5 | 0.7 | 8.2 | 1.1 | 72.0 | 5.1 | 10 | 65.2 | 5.5 | 664.1 | 18.1 | 71.2 | 6.6 | 8.5 | 0.8 | 70.8 | 7.5 | 0 |
| Parity 10 | 1024 | 10 | 2 | 96.1 | 3.9 | 1.0 | 0.0 | 6.8 | 0.4 | 97.6 | 0.8 | 1.0 | 0.0 | 7.5 | 0.7 | 51.3 | 4.9 | 8 | 44.2 | 5.7 | 921.2 | 1.0 | 80.7 | 3.4 | 20.0 | 0.0 | 97.6 | 1.3 | 8 |
| Parity 8 | 256 | 8 | 2 | 94.5 | 5.0 | 1.0 | 0.0 | 6.3 | 0.6 | 96.9 | 3.8 | 1.0 | 0.0 | 6.7 | 0.5 | 52.0 | 8.0 | 9 | 35.4 | 11.9 | 119.3 | 1.7 | 100.0 | 17.0 | 0.0 | 0.0 | 96.1 | 4.8 | 10 |
| Voting | 435 | 16 | 2 | 95.2 | 3.5 | 1.0 | 0.0 | 2.0 | 0.0 | 90.6 | 5.5 | 1.3 | 0.5 | 2.5 | 0.9 | 93.8 | 2.7 | 5 | 95.9 | 2.4 | 57.0 | 10.6 | 93.3 | 3.2 | 4.6 | 2.5 | 94.0 | 3.5 | 0 |
| Wine | 178 | 13 | 3 | 96.0 | 5.7 | 1.9 | 0.3 | 3.1 | 0.3 | 94.9 | 4.7 | 2.0 | 0.0 | 3.0 | 0.0 | 97.7 | 2.8 | 4 | 96.6 | 2.9 | 63.7 | 8.0 | 95.0 | 4.1 | 6.2 | 3.5 | 98.3 | 2.7 | 0 |
Breast Cancer Wisconsin and L. Breast data without vectors with missing values #P number of projections (hidden nodes) #K number of prototypes (output nodes) or number of nearest neighbours (kNN) #SV number of support vectors #N number of hidden nodes (MLP)
MLP Szczegółowe wyniki
LVQ Szczegółowe wyniki
MLP Matlab Toolbox
SNNS na binarnych
MLP porownanie rozbierznosci wynikow
MLP Szczegółowe wyniki (z kara)
Settings
Test
10 CV, stratified Data transformation: normalization
QPC/LVQ
Network parameters (Matlab)
Method lvq+qpc Initialization 5 LVQ learnign rate 0.010000 Attraction force 0.050000 Max. projections 5 K range 2 10 20 Attraction step 100 Precision (eps) 0.030000
QPC parameters (default)
lrate = 0.1; % learning rate maxiter = 1000; % MAX number of iterations init = 5; % number of initializations function = gauss; % width = 0.1; % function width eps = 0.001; % precision
PCA/LVQ
Network parameters (Matlab)
Method lvq+pca Initialization 5 LVQ learnign rate 0.010000 Attraction force 0.050000 Max. projections 5 K range 2 10 20 Attraction step 10 Precision (eps) 0.030000
LVQ1
Initialization 5 LVQ learnign rate 0.010000 K range 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Precision (eps) 0.010000
SVM
Ghost Miner 3.0 beta5 (21-05-2004)
Gaussian kernel Auto C & Bias (5CV, stratified)
MLP
C++ implementation
transfer function : sigmoid optimization : back propagation initiations : 10 hidden nodes : from 0 to 20
kNN
Ghost Miner 3.0 beta5 (21-05-2004)
Auto K (5CV, stratified) Euclidean distance measure Max K=10