Figure 3.
Modelling results of twenty times robust tests by four methods. Adaboost combined with NN outperformed the other modelling methods on the robustness testing. R2 is the coefficient of determination of training and test, respectively; RMSE is the root mean relative error. Different letters (a, b and c) indicated the significantly different of modelling methods by the least significant difference (LSD) multiple range test (p < 0.05) in SPSS 18.0 software.