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

Decision tree classifier Operational accuracy 0.84 0.84 balanced accuracy 0.62 0.60 precision 0.38 0.35 recall 0.32 0.27 F1 0.35 0.30 Table 5: Indices calculated based on the confusion matrix to quantify performance of the decision tree classifier, for the test and train set, and the operation model. This is the first model which is capable of correctly predicting more rock falls than the operational model with 18 additional correct predictions. Those predictions come with an additional 6 days when the model incorrectly predicted that there would be an event. This means 6 additional days the road could be operational with no accidents. Even though this model is safer for drivers the precision is greater than the operational. For future models this balance of weights will be adopted. Additionally the tuning will be done solely with the balanced accuracy as it is strongly correlated with the . β=1 Bagging Decision tree Classifier As described in the previous section, this method is an optimization of decision tree approach by using a combination of trees in order to predict if a rock will detach on a certain date. For this model both hyper-parameters tuned for the decision tree will be optimized once again and additionally the number of estimators will also be tuned. In order to visualise the interaction between the three variables 3 distinct heat maps will be represented. This enables the choice of the hyper-parameters that will be maintained. 39