Comparing Algorithm Performance in Machine Learning for Landslide Susceptibility Studies: An Overview

Muhammad  Arslan; Muhammad  Mubeen; G. Anandhi

Abstract

Machine learning algorithms have gained popularity over the past years, and with this comes the rise in their efficiencies; this study demonstrated that machine learning algorithms could successfully be used to identify landslide susceptibility areas. Machine learning algorithms like decision trees, support vector machines, K-Means, Hierarchical Clustering, Self-Organizing Maps (SOMs), artificial neural networks, and Bayesian networks have been used in recent studies. This study thus reviews several studies and results from several datasets and the advantages and disadvantages of these machine learning algorithms comprehensively; it is clear that the decision tree algorithm was the most accurate and interpretable algorithm tested and should be considered for further investigations into landslide susceptibility. The SVM algorithm also performed well in accuracy but was not as interpretable as the decision tree. The K-Means, Hierarchical Clustering, SOM, ANN, and Bayesian network algorithms all produced good results but were more complex and less interpretable than the decision tree. With these insights, future studies should focus on developing more sophisticated algorithms that can better capture the complexity of the problem. Additionally, more data should be collected to accurately reflect the area of interest's characteristics. Finally, the evaluation of algorithms should be conducted in a rigorous, systematic manner to ensure that the algorithms accurately capture the underlying relationships between landslide susceptibility and the various factors.

Abstract

Keywords