Mapping Landslide Susceptibility Using the Random Forest and Land Use Correlation in Northern Bandung Denny Lumban Raja (a*), Sukristiyanti (b), Yukni Arifianti (c), Fitriani Agustin(d), Roni Marudut Situmorang (e)
a) Study Program of Geological Technology, Bandung Polytechnic of Energy and Mining, Indonesia
b) Research Center for Geological Disaster, National Research and Innovation Agency (BRIN), Indonesia
c) Geological Agency of Indonesia, Indonesia
Abstract
Northern Bandung has a very high potential for landslides, triggered by its steep slopes and high elevations. Identifying landslides area is essential for evaluating and mitigating risks. This study conducted a classification of landslide susceptibility mapping in Northern Bandung using a machine learning technique called random forest. Machine learning methods are being used more frequently to solve various scientific and engineering problems. A total of 3056 landslide points were collected through field surveys, stakeholders from the Center for Volcanology and Geological Hazard Mitigation and Google Earth time series image interpretation. Furthermore, 14 factors potentially influencing landslides were considered, including slope gradient, elevation, aspect, profile curvature, flow direction, TRI, TWI, lithology, land use, road density, river density, lineament density, rainfall, and (Normalized Difference Vegetation Index (NDVI). This dataset was used to develop a geospatial database, with the landslide inventory subsequently split into 70% for training and 30% for testing the models. This approach validates the effectiveness of the Random Forest algorithm in accurately mapping landslide susceptibility, as demonstrated by an accuracy value of 0.98. The results indicate that areas most susceptible to landslides are correlated with dryland agriculture and plantations. These findings can support the development of effective strategies for landslide hazard mitigation
Keywords: Landslide- Machine learning- Random forest- Northern Bandung
