Open AccessLandslides Risk Prediction Using Cascade Neural Networks Model at Muş In Turkey
Author(s) -
Sohaib K. M. Abujayyab,
Azlan Saleh
Publication year - 2020
Publication title -
iop conference series. earth and environmental science
Language(s) - English
Resource type - Journals
eISSN - 1755-1307
pISSN - 1755-1315
DOI - 10.1088/1755-1315/540/1/012081
Subject(s) - landslide , artificial neural network , cascade , computer science , receiver operating characteristic , data mining , python (programming language) , software , scale (ratio) , matlab , artificial intelligence , machine learning , cartography , geology , geography , engineering , geotechnical engineering , chemical engineering , programming language , operating system
Globally, landslides risk represent a challenging issue that negatively affecting the infrastructure and human and neutral life. Among the former studies, the methods of predicting landslides risk maps found to be need further experiments. The aim of this study was to predict the landslide risk map at Muş in Turkey using cascade neural networks model. In this article, Trainlm function used to train 9954 sample points in the dataset using Matlab software. ArcGIS employed to prepare the explanatory variables, inventory landslides map, data sampling and producing the final landslides risk map. The developed model achieved the best performance accuracy by implanting an optimizer for the used number of neurons. After 60 training experiments, 52 neurons found the best number in this model. Chunks computing using Python programing in ArcGIS implemented to solve the intensive computing and data restructuring issues. Although the implementation at regional scale with 14015 km2, the final landslides risk map was successfully produced. The best-achieved performance accuracy was 80% based on receiver operating characteristic curve (ROC) and area under the curve (AUC). To summarize, the cascade neural networks model can reliably be implement in predicting landslides risk maps at regional-scale with the aid of chunks computing.