RINA - Démonstrateur d’utilisation de l’intelligence artificielle pour la gestion opérationnelle des risques naturels gravitaires géologiques | FEREC 2020

Figure 27: heat mapping of the performance of decision tree model as the hyperparameters varie for models with classes with same weights. Performance of model measured by (a) (b) β=1 balanced accuracy Observing both heat maps (Figure 27) it is visible that the and the balanced β=1 accuracy are related. This is logical as both indices should increase as the model improves and is capable of correctly predicting more rock falls. The model improves as the maximum depth decreases. This indicates the trees are overfitting and searching to fit the model too closely. This is especially true when the min sample leaf is smaller facilitating the overfitting. The optimum model would be a small tree with a maximum depth of 3 and a minimum number of samples per leaf of 8. This model was created and the confusion matrix is presented in figure 28. Actual 0 1 Predicted 0 909 114 1 35 32 Figure 28: Confusion matrix representing the result of predictions based on the decision tree method for the test set 36