Modelling Soil Water Retention Using Support Vector Machines with Genetic Algorithm Optimisation
Table 3
Developed models performance comparisons.
Potential (kPa)
Model type
Number of SV
Training dataset
Testing dataset
RMSE
RMSE
Aim 1
Aim 2
Aim 1
Aim 2
Aim 1
Aim 2
Aim 1
Aim 2
Aim 1
Aim 2
−0.98
C-linear
137.2
179.1
0.92
0.92
0.0184
0.0185
0.90
0.90
0.0194
0.0193
(13.71)
(6.66)
(0.006)
(0.006)
(0.0005)
(0.0005)
nu-linear
189.8
189.8
0.92
0.92
0.0186
0.0186
0.90
0.90
0.0194
0.0194
(1.99)
(1.81)
(0.006)
(0.006)
(0.0005)
(0.0005)
C-radial
359.5
187.1
0.99
0.97
0.0051
0.0116
0.60
0.77
0.0404
0.0293
(5.38)
(1.73)
(0.001)
(0.001)
(0.0003)
(0.0002)
nu-radial
372.3
198.4
0.99
0.94
0.0052
0.0160
0.59
0.92
0.0414
0.0175
(3.13)
(4.58)
(0.001)
(0.006)
(0.0003)
(0.0005)
−3.1
C-linear
92.5
117.7
0.62
0.62
0.0485
0.0497
0.67
0.68
0.0384
0.0391
(7.11)
(85.18)
(0.010)
(0.008)
(0.0008)
(0.0011)
nu-linear
189.1
189.2
0.61
0.61
0.0497
0.0497
0.68
0.68
0.0363
0.0363
(1.45)
(1.81)
(0.011)
(0.011)
(0.0009)
(0.0009)
C-radial
352.1
187.4
0.92
0.84
0.0225
0.0312
0.23
0.48
0.0735
0.0492
(10.79)
(1.35)
(0.007)
(0.012)
(0.0008)
(0.0009)
nu-radial
372.4
194.5
0.92
0.68
0.0227
0.0448
0.22
0.70
0.0770
0.0348
(2.80)
(2.88)
(0.007)
(0.018)
(0.0007)
(0.0010)
−9.81
C-linear
134.0
133.7
0.76
0.75
0.0516
0.0529
0.72
0.72
0.0497
0.0499
(14.34)
(111.77)
(0.004)
(0.011)
(0.0007)
(0.0016)
nu-linear
189.7
189.7
0.76
0.76
0.0517
0.0517
0.72
0.72
0.0498
0.0498
(1.49)
(1.77)
(0.004)
(0.004)
(0.0007)
(0.0007)
C-radial
360.2
187.7
0.98
0.93
0.0151
0.0282
0.35
0.63
0.0959
0.0577
(8.46)
(1.77)
(0.003)
(0.003)
(0.0014)
(0.0006)
nu-radial
372.4
194.8
0.98
0.85
0.0152
0.0416
0.33
0.82
0.0985
0.0395
(2.95)
(2.49)
(0.004)
(0.007)
(0.0015)
(0.0010)
−31.02
C-linear
141.1
123.1
0.77
0.77
0.0516
0.0528
0.71
0.70
0.0528
0.0532
(6.90)
(102.16)
(0.004)
(0.008)
(0.0006)
(0.0012)
nu-linear
190.0
189.9
0.77
0.77
0.0522
0.0522
0.71
0.71
0.0532
0.0532
(2.05)
(2.42)
(0.004)
(0.004)
(0.0006)
(0.0006)
C-radial
351.2
187.8
0.99
0.93
0.0123
0.0292
0.36
0.67
0.0957
0.0554
(11.31)
(2.20)
(0.001)
(0.005)
(0.0007)
(0.0009)
nu-radial
372.3
195.6
0.99
0.85
0.0126
0.0419
0.34
0.80
0.0998
0.0436
(3.06)
(4.03)
(0.002)
(0.007)
(0.0009)
(0.0007)
−491.66
C-linear
109.0
80.8
0.72
0.71
0.0502
0.0518
0.67
0.65
0.0495
0.0517
(17.95)
(54.42)
(0.008)
(0.016)
(0.0013)
(0.0023)
nu-linear
189.3
189.7
0.71
0.71
0.0508
0.0509
0.67
0.67
0.0493
0.0493
(1.77)
(1.64)
(0.009)
(0.009)
(0.0015)
(0.0015)
C-radial
364.8
186.9
0.99
0.93
0.0091
0.0252
0.36
0.60
0.0875
0.0569
(7.64)
(2.51)
(0.002)
(0.004)
(0.0011)
(0.0006)
nu-radial
372.5
202.1
0.99
0.80
0.0092
0.0421
0.35
0.70
0.0890
0.0475
(3.10)
(8.35)
(0.002)
(0.020)
(0.0011)
(0.0026)
−1554.78
C-linear
92.7
136.1
0.69
0.68
0.0459
0.0470
0.63
0.63
0.0468
0.0465
(21.16)
(77.41)
(0.012)
(0.013)
(0.0016)
(0.0014)
nu-linear
189.4
189.7
0.68
0.68
0.0464
0.0464
0.64
0.64
0.0463
0.0463
(1.90)
(2.21)
(0.014)
(0.014)
(0.0017)
(0.0017)
C-radial
364.6
187.7
0.99
0.92
0.0093
0.0232
0.32
0.58
0.0794
0.0511
(5.62)
(2.21)
(0.002)
(0.003)
(0.0008)
(0.0008)
nu-radial
372.5
199.9
0.99
0.77
0.0094
0.0396
0.32
0.67
0.0805
0.0446
(2.95)
(7.49)
(0.002)
(0.025)
(0.0008)
(0.0026)
The presented values are averages of the other ten -fold submodels. In the case of the number of support vectors, RMSE and for the training dataset and values in brackets are standard deviations. Columns described by “aim 1” present data for models developed using RMSE as the aim function. Label “aim 2” is linked with models developed using proposed new form of the aim function.
