Flood Vulnerability Analysis using Random Forest Method in Gresik Regency, Indonesia

Abstract

Recently, machine learning models have been used for flood vulnerability mapping. The purpose of this study is analyze the level of flood vulnerability in Gresik Regency using machine learning Random Forest. Data processing is done by dividing the flood occurrence dataset that contains parameter values into two datasets, training data as input data for random forest and testing data to evaluate the output model. The results of the flood vulnerability level are divided into 3 classes of flood vulnerability, which are low, medium and high levels vulnerability. The results of the importance degree index show that the most influential parameter of flood vulnerability is the river with a value of 41.025%. Furthermore, The performance of the random forest model shows an overall accuracy and kappa coefficient of 0.859 and 0.718, respectively while the evaluation using the ROC curve shows an AUC value of 0.97 which indicates the classification results are included in the very good category.