Landslide Susceptibility Mapping with Stacking Ensemble Machine Learning

• Main Authors: Solihin M.I., Yanto, Hayder G., Maarif H.A.-Q.
• Other Authors: 16644075500
• Format: Conference Paper
• Published: Springer Nature 2024
• Subjects: Imbalanced classification; Landslide susceptibility mapping; Machine learning algorithms; Stacking ensemble

Landslide susceptibility mapping (LSM) is an important preliminary effort to reduce the risk and harshness of landslide disasters. While numerous methods have been proposed, machine learning (ML) is the most popular approach that has been applied across the globe. One of the prominent methods to improve machine learning accuracy is by using ensemble method which basically employs multiple base models. In this paper, the stacking ensemble method is used to increase the accuracy of the machine learning model for LSM where the base (first-level) learners use five ML algorithms namely decision tree (DT), k-nearest neighbor (KNN), AdaBoost, extreme gradient boosting (XGB) and random forest (RF). The second-level learner uses logistic regression (LR) to aggregate the final prediction output. The landslide data together with its conditioning factors (feature variables) collected from three districts in the Central Java Province, Indonesia, has been used as the case study for the LSM. As the data are extremely imbalanced, Adaptive Synthetic (ADASYN) resampling technique was picked to balance the data between two classes, i.e., landslide and non-landslide. This is because the occurrence of non-slide incidents is much more than the landslide. The evaluation results of the LSM performance show that the proposed stacking ensemble ML improves the overall accuracy of the individual base ML model even when it is compared with RF which is naturally also ensemble ML. � 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.