|
Evaluation of the logistic regression model for predicting landslide-prone areas; case study: Qazvin county
|
Rouholla Jafarisirizi *  , Mohammad Hassan Khamesi-Meybodi , Javad Sadidi  |
| Department of Remote Sensing and GIS, Faculty of Earth Sciences, Shahid Beheshti University, Tehran, Iran. |
|
|
Abstract: (54 Views) |
Objective: Landslides are recognized as a potential hazard. Identifying landslide-prone areas can improve landslide management and prediction. This potential hazard poses a serious threat to human infrastructure and even human life. Advances in spatial sciences and machine-learning algorithms have increased efforts to develop predictive maps of landslide occurrence.
Method: In this study, 10 criteria, including soil type, lithology, precipitation, distance from faults, vegetation cover, distance from streams, distance from urban areas, distance from rural settlements, slope, and elevation, were selected to develop a landslide occurrence prediction map for Qazvin County. A spatial database was established using GIS-based analyses. A spatial resolution and pixel size of 30 × 30 m were adopted for all criteria to produce the final prediction map at the same resolution. The landslide risk map was then developed using the logistic regression model and training data for landslide occurrence. Model performance was evaluated using precision–recall metrics and the area under the receiver operating characteristic (ROC) curve (AUC).
Results: The prediction map showed that areas classified as having high and very high landslide risk were predominantly distributed as linear and curvilinear belts across different parts of the study area. This spatial pattern may be influenced by fault alignments, drainage networks, lithological variations, and abrupt changes in slope. The results showed that, at lower thresholds, the initial positive cases were identified relatively well; however, as recall increased, precision declined to approximately 0.6–0.7. Meanwhile, the AUC was 0.557, which is close to random performance. Although the precision–recall analysis indicated slightly better model performance at certain thresholds, the ROC analysis confirmed that the model had generally low discriminatory power.
Conclusions: The results indicate that the logistic regression model and the selected criteria do not provide high predictive performance for identifying landslide risk. Future studies are therefore recommended to employ a wider range of machine-learning and deep-learning methods, along with more diverse and comprehensive datasets, to improve the prediction of landslide-prone areas.
|
|
| Keywords: Landslide, Logistic Regression, Risk Prediction, GIS, Qazvin County. |
|
|
Full-Text [PDF 1220 kb]
(25 Downloads)
|
Type of Study: Applicable |
Subject:
Special
Received: 2025/11/28 | Accepted: 2026/08/9 | Published: 2026/08/9
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
