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

Actual 0 1 Predicted 0 917 116 1 27 30 Figure 26: Confusion matrix representing the result of predictions based on the nearest neighbors method for the test set. accuracy 0.87 0.84 balanced accuracy 0.59 0.60 precision 0.53 0.35 recall 0.21 0.27 F1 0.30 0.30 Table 3: Indices calculated based on the confusion matrix to quantify performance of the nearest neighbors model and the operation model. 5.3. Decision trees For this model, more than one parameter are tuned in order to optimise the results. A heat map was plotted to observe how the maximum depth and the minimum samples per leaf would affect the results. The score and the balanced accuracy are calculated for each combination β=1 and a heat map is created (Figure 27). This image permits the choice of the combination of parameters that optimises the result. It is important to note that both indices are only calculated in respect to the test set (a grid search is made) and it must be later verified if overfitting does occur. 35