TABLE I.
Results for KRR. Mean RMSE and standard deviation (in parentheses).
| Kernel | RBF | Polynomial | ||||||
|---|---|---|---|---|---|---|---|---|
| Data set/Method | Modified ε-constraint | ε-constraint | CV | GCV | Modified ε-constraint | ε-constraint | CV | GCV |
| Housing | 2.89*(0.77) | 3.01(0.78) | 3.25(0.84) | 4.01(1.01) | 3.71(0.87) | 4.38(0.99) | 4.24(1.03) | 8.67(6.78) |
| Mpg | 2.51*(0.52) | 2.59*(0.57) | 2.72*(0.40) | 2.61(0.52) | 2.82(0.45) | 3.25(0.58) | 3.24(0.57) | 3.21(0.80) |
| Slump | 6.62*(1.49) | 7.36(2.29) | 6.70(1.53) | 22.1(8.95) | 7.09(1.22) | 8.85(2.05) | 9.86(1.53) | 7.20(1.77) |
| Price | 2.21*(0.90) | 2.73(1.54) | 2.42(0.90) | 8.88(5.43) | 3.08(1.20) | 3.29(1.50) | 4.01(1.48) | 3.41(1.5) |
| Diabetes | 0.55*(0.23) | 0.72(0.33) | 0.57(0.19) | 0.88(0.31) | 0.52*(0.17) | 0.60(0.20) | 2.31(0.87) | 0.62(0.33) |
| Wdbc | 31.46*(1.59) | 32.15*(4.86) | 31.50*(4.37) | 50.30(9.13) | 34.11(4.23) | 35.12(5.21) | 46.61(6.89) | 32.04*(4.35) |
| Servo | 0.51*(0.29) | 0.56(0.30) | 0.59(0.32) | 0.81(0.52) | 0.70(0.25) | 0.70(0.25) | 0.75(0.25) | 0.65(0.27) |
| Puma-8nm | 1.44*(0.02) | 1.51(0.03) | 2.42(0.05) | 1.44*(0.03) | 1.42*(0.02) | 1.89(0.04) | 1.89(0.04) | 1.46(0.02) |
| Puma-8nh | 3.65(0.03) | 3.64(0.03) | 3.98(0.06) | 3.56*(0.04) | 5.08(1.26) | 5.28(0.19) | 4.11(0.14) | 3.61*(0.06) |
| Puma-8fm | 1.13*(0.01) | 1.19(0.02) | 1.19(0.01) | 1.14*(0.02) | 1.27(0.01) | 1.37(0.09) | 1.29(0.005) | 1.27(0.01) |
| Puma-8fh | 3.23*(0.01) | 3.45(0.02) | 3.23*(0.01) | 3.23*(0.01) | 3.78(0.16) | 4.86(0.14) | 3.23(0.01) | 3.24*(0.02) |
| Kin-8nm | 0.11*(0.002) | 0.15(0.003) | 0.16(0.002) | 0.19(0.02) | 0.18(0.0008) | 0.24(0.03) | 0.22(0.002) | 0.19(0.01) |
| Kin-8nh | 0.18*(0.001) | 0.18(0.002) | 0.19(0.002) | 0.18(0.002) | 0.20(0.002) | 0.29(0.006) | 0.24(0.003) | 0.22(0.003) |
| Kin-8fm | 0.016(0.002) | 0.016(0.03) | 0.013*(0.0001) | 0.339(0.202) | 0.013*(0.0001) | 0.02(0.0001) | 0.16(0.003) | 0.015(0.0001) |
| Kin-8fh | 0.07(0.002) | 0.061(0.002) | 0.043*(0.0002) | 0.043*(0.0002) | 0.046(0.0002) | 0.046(0.0002) | 0.16(0.003) | 0.050(0.0003) |
In each kernel, the best result is in bold.
*
The symbol is used to indicate the top result over all methods and kernels.