Assessment of Landslide Susceptibility Mapping Using Artificial Bee Colony Algorithm Based on Different Normalizations and Dimension Reduction Techniques

• Published in: Arabian journal for science and engineering (2011) 2022-06, Vol.47 (6), p.7243-7260
• Main Authors: Huqqani, Ilyas Ahmad, Tay, Lea Tien, Mohamad-Saleh, Junita
• Format: Article
• Language: English
• Subjects: Algorithms; Engineering; Humanities and Social Sciences; Landslides; Mapping; Mean; Model accuracy; multidisciplinary; Principal components analysis; Reduction; Research Article-Earth Sciences; Risk management; Science; Search algorithms; Swarm intelligence
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-021-06013-8

In this paper, the artificial bee colony (ABC) algorithm is used to generate landslide susceptibility maps. To the best of our knowledge, it is the first time the ABC algorithm is applied as a classifier on landslide susceptibility mapping. Twelve features are considered as inputs for the ABC model in this research work. These features contain redundant and correlated information. Therefore, normalization and dimension reduction are applied in the pre-processing stage. Three different normalization methods are investigated, i.e. minimum and maximum (Min-Max), mean standard deviation (Mean-SD), and median and interquartile range (Med-IQR). Two dimensionality reduction techniques, i.e., principal component analysis (PCA) and canonical variate analysis (CVA), are utilized to deal with high-dimensional information and eliminate correlated information in the data. In total, six different schemes of input data based on the three normalization methods and two techniques of dimension reduction are applied to train the ABC model. There are a total of 382 landslide incidents occurred at various locations on Penang Island, out of which 255 are selected for training and remaining 127 are used for the validation purpose of the model. The model classification accuracies and receiver of characteristics (ROC) curve are applied to evaluate the model’s performance. The results show that the ABC model trained using Min-Max normalization and CVA scheme generates the best results among these six schemes. The landslide susceptibility maps generated in this research offer significant information for landslide risk management.