Makine Öğrenmesi Teknikleri Kullanılarak Heyelan Duyarlılık Haritalarının Üretilmesi: İnegöl (Bursa) Örneği

Abstract

In this thesis study, a research was conducted to determine landslide susceptibility in the İnegöl district of Bursa province. 13 environmental factors influencing landslide occurrence (slope, aspect, elevation, lithology, soil group, NDVI, land use, precipitation, distance to fault lines, distance to streams, distance to roads, curvature, and topographic wetness index) were analyzed within a Geographic Information System (GIS) environment; each factor was reclassified into five classes and thematic suitability maps were generated. The dominant class for each factor in the study area was identified to assemble the input dataset for the classification algorithms. Four different machine learning algorithms (Logistic Regression, Support Vector Machines, LightGBM, and Random Forest) were applied in a comparative framework. Model performance was assessed using metrics such as accuracy, precision, recall, and F1 score. According to the analysis results, the Random Forest algorithm achieved the highest performance with 98% accuracy, 98% precision, 98% recall, and a 98% F1 score. The study demonstrates that Random Forest delivers high performance in spatial data analyses and, when integrated with GIS, provides a reliable approach for landslide risk mapping. The resulting susceptibility map can serve as a decision-support tool for disaster risk reduction and land-use planning.