Flood Prediction Using Machine Learning Model Integrated with Geographical Information System

Flooding in Indonesia is still a frequent natural disaster compared to other types of disasters. In addition, the number of flood events also shows an increase every year. This research aims to develop a flood prediction model as a preventive measure as an early warning system and flood risk mitigation management that may occur based on Geographical Information System (GIS). It is expected that areas that have the potential to experience flooding can be more proactive in making preparations before flooding. This prediction model uses a classification type machine learning (ML) algorithm with training data involving rainfall within 12 months. The model evaluation results use two techniques: confusion matrix and K-Fold cross validation and each fold is calculated for accuracy. The K-Nearest Neighbors (KNN) model with a value of K = 31 gets the highest accuracy value of 88.89%, Decision Tree (DT) of 72.22%, and Naive Bayes of 78%. The average accuracy using K-Fold resulted in 89.09% for KNN, 77.12% for DT, and 86.59% for Naive Bayes. By considering these results, this research chose the KNN method to be applied in the prediction model. The code was rewritten in the Flask framework to be used as an API and integrated with Laravel as a Backend platform and Frontend using Bootstrap, JQuery, Axios, and LeafletJS as map visualisation. With this research, it is hoped that it can be one of the solutions in predicting as well as early warning of floods so that it can provide sufficient time for affected residents to make preparations for flooding.