3. Results and Discussion
The 5-fold cross validation indicates that TAN model has the lowest RMSE (Table 1). Friedman’s test reports significant differences (p-value < 0.05) among the tested models in each dataset. The post-hoc test shows significant differences between the TAN model and both MLP and LR models, with the former having significantly lower RMSE than the latter models (Fig. 1). According to the statistical test applied, the TAN model was not outperformed by any other model in terms of RMSE. This means that, from the point of view of accuracy, the TAN is competitive with other popular methods used for regression. The error maps (Fig. 2) provide detailed information about the reliability of the results in every single watershed. It is worth noting that northern areas present low errors on the 4 maps, while watersheds in the center show higher errors, especially on the MLP map. These maps emphasize that both TAN and M5P show lower error dispersion and, therefore, are more accurate in the whole study area. Besides accuracy, BNs provide a number of advantages, including the capability of predicting the posterior density function of the response variable given some evidence, which is more useful and accurate than a single mean value.
3. Results and Discussion The 5-fold cross validation indicates that TAN model has the lowest RMSE (Table 1). Friedman’s test reports significant differences (p-value < 0.05) among the tested models in each dataset. The post-hoc test shows significant differences between the TAN model and both MLP and LR models, with the former having significantly lower RMSE than the latter models (Fig. 1). According to the statistical test applied, the TAN model was not outperformed by any other model in terms of RMSE. This means that, from the point of view of accuracy, the TAN is competitive with other popular methods used for regression. The error maps (Fig. 2) provide detailed information about the reliability of the results in every single watershed. It is worth noting that northern areas present low errors on the 4 maps, while watersheds in the center show higher errors, especially on the MLP map. These maps emphasize that both TAN and M5P show lower error dispersion and, therefore, are more accurate in the whole study area. Besides accuracy, BNs provide a number of advantages, including the capability of predicting the posterior density function of the response variable given some evidence, which is more useful and accurate than a single mean value.
การแปล กรุณารอสักครู่..
