Integrating rule-based algorithms with fuzzy rule induction on regional landslide susceptibility modeling

This study integrates Decision Tree (DT) and Particle Swarm Optimization (PSO) algorithms with Fuzzy Rule Induction (FRI) operator respectively (called DT-FRI and PSO-FRI) to assess landslide susceptibility according to existing rainfall-induced and shallow landslide events. The constructed landslide susceptibility models are applied to classify and verify occurrence samples. In this study, two strategies are applied for the model verification, i.e. space- and time-robustness. The former is to separate samples into training and check data based on a single event. The latter is to predict (classify) later landslide events with a landslide susceptibility model which is constructed from earlier events. Eleven geospatial factors are considered, including topographic, vegetative, environmental, geological and man-made information. The landslide inventory and factors are overlapped to obtain the training and check data for modeling and verification. Experimental results show that applying the conventional DT algorithm can reach high modeling accuracy respectively based on the space-robustness strategy but both have poor performance to predict (classify) consequent events (time-robustness). After integrating with FRI, the prediction (classification) results are significantly improved, especially using PSO-FRI models.