Comparative assessing the performance of ANN, RF and CNN machine learning methods in identifying landslide prone areas

Landslides are one of the natural hazards in mountainous areas that cause a lot of damage every year, thus, determining the landslide prone area is very important. This study assessed the performance of artificial neural network (ANN), Random Forest (RF), and Convolutional Neural Network (CNN) methods in identifying landslide prone areas. The study area is Lorestan Province, Khorramabad Watershed in western Iran, a region highly susceptible to landslides. After pre-processing the satellite images, the training samples were collected using field visits. Then, the neural network with a modifed structure was used for classification based on the simultaneous integration of the algorithm used. The available data were divided into ۷۰% for the training, and ۳۰ % for the validation stages. The performance of the generated classification maps of three employed methods were evaluated using the overall accuracy and confusion matrix. The results of evaluating the performance and accuracy of the CNN algorithm for identifying landslide areas show ۹۳% overall accuracy. While the evaluation of the results obtained from ANN and RF methods shows that the overall accuracy of the neural network method is ۹۰% and its overall accuracy is ۸۸% and in the random forest method the overall accuracy is ۸۴% and the overall accuracy is ۸۲%; This study shows that the proposed method has shown the best performance compared to other methods according to evaluation criteria.These findings highlight the superiority of the CNN-based approach in accurately mapping landslide-prone areas, making it a reliable tool for future hazard assessment and risk management in mountainous regions.

کلیدواژه ها:

Landslide occurrence ، Machine Learning ، Convolutional Neural Network ، Landslide susceptibility mapping ، Lorestan province