A Comparative Analysis of the Machine Learning Techniques for Finding the Relationship Between Soil and Its Appropriate Sub Parameters in Landslide Susceptibility Mapping

Abstract

Landslides are known to be one of the most common natural disaster in hilly areas. These landslides have caused a huge loss to both property and life [2]. Hence, there arises a need for adoption of proper measures for minimizing the risk of landslides by zoning the areas as per its vulnerability to landslides.

Zonation of landslides can be done using a landslide susceptibility Map[LSM]. Thus, there arises the need for an efficient model for the development of an LSM.  Landslides causal factors are very imperative in determining the accuracy of an LSM model as these factors will be considered as input parameters for the model.   Some parameters may have its own sub parameters, which may be taken into consideration for the development of a model. Hence, finding the relationship between parameter and its sub parameters may be helpful in the development of an efficient and effective model for an LSM.  The role of various machine learning techniques has become very vital in the area of geotechnical applications. The main contribution of the paper is to make a comparative analysis of various machine learning techniques to find the relevance between the main parameter and its sub parameter. Here, soil has been considered as the main parameter.  The techniques used are Word2Vec, Sequence Matcher and Term Frequency-Inverse Document Frequency (TF-IDF). The correlation between the parameter and its sub parameter is based on the ratio obtained between the sub parameters and input parameter. The study showed that SequenceMatcher showed better results as compared to Word2Vec and TF-IDF.