search

sam-mata / antarctic-ai

Researching evolutionary learning methods to predict changes in Antarctic ice sheets.
MIT License
1 stars 0 forks source link

Re-Evaluate Higher Explainability Models #74

Closed sam-mata closed 1 month ago

sam-mata commented 1 month ago

Details

Several models initially excluded should now be re-evaluated due to the higher degree of explainability and interpretability inherent in their architectures, compared to currently higher performing alternatives.

Models

sam-mata commented 3 weeks ago

Broad Results

Ice Thickness

Model MSE RMSE MAE R2
RandomForestRegressor 0.000772 0.027790 0.009127 0.988045
ExtraTreesRegressor 0.000792 0.028142 0.010056 0.987739
DecisionTreeRegressor 0.001930 0.043930 0.009419 0.970125
AdaBoostRegressor 0.006729 0.082032 0.043229 0.895826
Linear Regression 0.007701 0.087756 0.060153 0.880783
Ridge 0.007701 0.087755 0.060152 0.880783
Lars 0.007701 0.087756 0.060153 0.880783
BayesianRidge 0.007701 0.087756 0.060153 0.880783
ARDRegression 0.008293 0.091065 0.063821 0.871622
HuberRegressor 0.008525 0.092330 0.061983 0.868030
TheilSenRegressor 0.009081 0.095296 0.063195 0.859414
KNeighborsRegressor 0.010773 0.103795 0.045649 0.833221
OrthogonalMatchingPursuit 0.011450 0.107004 0.078276 0.822749
XGBoost 0.011933 0.109238 0.091619 0.815272
LightGBM 0.012103 0.110014 0.092334 0.812638
PLSRegression 0.013589 0.116574 0.080526 0.789626
GradientBoostingRegressor 0.013929 0.118020 0.096844 0.784374
LinearSVR 0.021850 0.147817 0.112234 0.661750
ElasticNet 0.053446 0.231185 0.197646 0.172615
Lasso 0.064599 0.254163 0.217919 -0.000034
LassoLars 0.064599 0.254163 0.217919 -0.000034
PassiveAggressiveRegressor 0.065693 0.256306 0.190243 -0.016971
RANSACRegressor 0.093425 0.305655 0.169790 -0.446284
MLPRegressor 0.134074 0.366162 0.307961 -1.075557
SGDRegressor 1.953786e+29 4.420165e+14 3.996029e+14 -3.024589e+30

Ice Velocity

Model MSE RMSE MAE R2
ExtraTreesRegressor 0.000083 0.009127 0.001184 0.485918
RandomForestRegressor 0.000100 0.009984 0.001382 0.384788
LightGBM 0.000119 0.010890 0.002545 0.268137
GradientBoostingRegressor 0.000145 0.012038 0.002800 0.105600
Linear Regression 0.000157 0.012547 0.003300 0.028451
Ridge 0.000157 0.012547 0.003300 0.028451
BayesianRidge 0.000157 0.012547 0.003298 0.028455
AdaBoostRegressor 0.000157 0.012543 0.003245 0.029123
Lars 0.000158 0.012552 0.003316 0.027709
PLSRegression 0.000158 0.012580 0.003260 0.023283
Lasso 0.000162 0.012730 0.003320 -0.000190
ElasticNet 0.000162 0.012730 0.003320 -0.000190
LassoLars 0.000162 0.012730 0.003320 -0.000190
OrthogonalMatchingPursuit 0.000162 0.012733 0.003240 -0.000570
ARDRegression 0.000162 0.012730 0.003320 -0.000190
LinearSVR 0.000163 0.012763 0.002880 -0.005251
KNeighborsRegressor 0.000164 0.012787 0.002510 -0.009148
TheilSenRegressor 0.000164 0.012826 0.001971 -0.015187
HuberRegressor 0.000165 0.012863 0.001947 -0.021068
RANSACRegressor 0.000166 0.012881 0.001971 -0.023980
XGBoost 0.000169 0.012999 0.002879 -0.042789
DecisionTreeRegressor 0.000264 0.016241 0.001200 -0.627897
PassiveAggressiveRegressor 0.009479 0.097361 0.096952 -57.501461
MLPRegressor 0.212440 0.460913 0.293794 -1310.085758
SGDRegressor 1.513591e+29 3.890490e+14 3.519026e+14 -9.341201e+32

Ice Mask

Model MSE RMSE MAE R2
ExtraTreesRegressor 0.003977 0.063061 0.015391 0.982640
RandomForestRegressor 0.004584 0.067703 0.013942 0.979989
DecisionTreeRegressor 0.009162 0.095716 0.013327 0.960005
AdaBoostRegressor 0.017709 0.133075 0.050483 0.922690
KNeighborsRegressor 0.020023 0.141503 0.042338 0.912587
LightGBM 0.040738 0.201835 0.187511 0.822156
XGBoost 0.041059 0.202630 0.188063 0.820753
Linear Regression 0.042598 0.206393 0.150208 0.814034
Ridge 0.042598 0.206393 0.150207 0.814034
BayesianRidge 0.042598 0.206393 0.150207 0.814034
ARDRegression 0.043074 0.207544 0.151993 0.811955
Lars 0.045447 0.213182 0.154638 0.801598
GradientBoostingRegressor 0.045505 0.213320 0.195905 0.801342
TheilSenRegressor 0.050466 0.224646 0.152986 0.779686
PLSRegression 0.050089 0.223805 0.161985 0.781333
HuberRegressor 0.051162 0.226190 0.152042 0.776648
LinearSVR 0.094719 0.307765 0.236112 0.586495
ElasticNet 0.111886 0.334493 0.311609 0.511552
OrthogonalMatchingPursuit 0.136478 0.369429 0.282073 0.404192
PassiveAggressiveRegressor 0.161796 0.402239 0.319807 0.293664
Lasso 0.191213 0.437279 0.421482 0.165243
LassoLars 0.191213 0.437279 0.421483 0.165240
MLPRegressor 0.271285 0.520850 0.449190 -0.184319
RANSACRegressor 0.380674 0.616988 0.389372 -0.661869
SGDRegressor 2.511256e+29 5.011244e+14 4.963325e+14 -1.096313e+30