Prediction of Flood Susceptibility in West Java, Indonesia Based on Geospatial Factor Analysis and Machine Learning Models

초록

In recent years, the urgency of flood management in West Java has increased due to the high frequency and intensity of flooding, especially in lowland areas connected to river basins, such as agricultural and urban areas. A comprehensive analysis of flood triggering factors is necessary to predict flood-prone areas more accurately. This study aims to predict flood risk in West Java by combining remote sensing methods, geospatial analysis, and machine learning approaches. A total of 114 historical flood locations that inundated several areas in West Java were identified. Flood inventory data was separated into 70% as training data and 30% as testing data. Flood-triggering factors, including topography, hydrology, environment, and climatology, were collected and selected based on a multicollinearity assessment and a frequency ratio (FR) analysis. Machine learning models using a convolutional neural network (CNN) and long short-term memory (LSTM) algorithms were trained with flood locations as the dependent variable and triggering factors as the independent variables. The results showed that the cities of Bandung, Bekasi, Sukabumi, and Cianjur are at high to very high susceptibility levels. The model's prediction performance was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC) analysis, where CNN with an AUC value of 0.720 was superior in conducting predictions compared to LSTM with an AUC value of 0.706. This research aligns with government policy and the Sustainable Development Goals and can be used as an effort to ensure public safety from flood disasters.

키워드