Evaluation of Tree-Based Artificial Intelligence Models to Predict Flood Risk using GIS

Floods are one of the most devastating types of natural disasters that every year causes the loss of human lives and properties around the world. The purpose of this study is to evaluate and compare the capability of three machine learning models namely Naïve Bayes Tree (NBTree), Alternating Decision Tree (ADTree), and Random Forest (RF) to predict flood risk in Maneh and Samalqan county. The novelty of the present study is the presentation of tree-based hybrid models that have been less used in previous research. To prepare a flood reference map in the study area, 300 flood-prone locations were identified and were divided into training and validation data sets through random selection with a ratio of 70 to 30. The spatial database of the flood was created using 15 hydrogeological and environmental criteria affecting the flood. Finally, three flood risk prediction maps were generated using NBTree, ADTree, and RF models. To validate the flood risk predicting models, the Area Under the Curve (AUC) factor and the statistical criteria of Positive predictive rate, negative predictive rate, sensitivity, specificity, and accuracy were used. The results showed that the RF model had higher accuracy than the NBTree and ADTree models in predicting flood risk in the study area. The results also showed that the risk of flooding in the central areas of the study area is higher than other areas due to lower altitude and slope.