<![CDATA[This study aims to map landslide susceptibility in Batu City, East Java Province, Indonesia, using machine learning on the Google Earth Engine platform. Five algorithms were used: random forest (RF), classification and regression trees, support vector machine, gradient tree boosting, and k-nearest neighbors. Landslide susceptibility data were collected from 276 sample points, comprising 138 landslides and 138 nonlandslides. The data were divided into two parts: 70% for training and 30% for testing. The conditioning factors initially considered consisted of 13 variables. Pearson correlation matrix analysis indicated that annual rainfall should be excluded to avoid multicollinearity, resulting in 12 predictors used for mapping: elevation, slope, aspect, landforms, distance from rivers, topographic wetness index, soil types, land use/land cover, normalized difference vegetation index, distance from roads, geological formations, and distance from lineaments. Model performance was evaluated using six metrics: accuracy, precision, recall (sensitivity), F1-score, kappa coefficient, and area under the receiver operating characteristic curve. Model validation results indicated that the RF algorithm was the best model. The RF-based landslide susceptibility map classified the area into five classes: very low class covering an area of 60.70 km2 (30.49%), low class covering an area of 29.47 km2 (14.80%), medium class covering an area of 16.21 km2 (8.14%), high class covering an area of 22.56 km2 (11.33%), and very high class covering an area of 70.16 km2 (35.24%). Variable importance analysis showed that slope is the most influential factor, with a mean decrease in impurity of 43.77 (15.36%), whereas soil types had the least contribution. This study successfully mapped landslide susceptibility using a machine learning approach, enabling direct support for disaster mitigation planning and risk-based land use in Batu City.]]>
Copyrights © 2026
