Table 4. Support vector machines.
Kernel Function | Linear | Quadratic | Polynomial | Gaussian Radial Basis function | Multilayer Perceptron Kernel | |||||
Method | Training | Test | Training | Test | Training | Test | Training | Test | Training | Test |
Least Squares | 35.43 | 54.75 | 0.68 | 44.08 | 0 | 45.85 | 0.21 | 49.08 | 35.58 | 49.61 |
Quadratic Programming | 35.36 | 54.74 | 0.01 | 44.22 | 0 | 44.50 | 0.18 | 48.93 | Fails to converge | |
Sequential Minimal Optimization | 33.38 | 55.87 | 0.88 | 45.31 | 0 | 45.01 | 1.38 | 49.67 | 62.57 | 56.85 |
Training set and test set errors are shown for each combination of kernel function and method used to find the separating hyperplane. Training set errors as low as 0% were achieved with kernel functions of higher degrees and complexities, but test set errors were always >45%, indicating model over-fitting.