Comparison of Moving-Average, Lazy, and Information Gain Methods for Predicting Weekly Slope-Movements: A Case-Study in Chamoli, India

Abstract

Landslide incidence is common in hilly areas. In particular, Tangni in Uttrakhand state between Pipalkoti and Joshimath has experienced a number of landslide incidents in the recent past. Thus, it is important to forecast slope-movements and associated landslide events in advance of their occurrence to avoid the associated risk. A recent approach to predicting slope-movements is by using machine-learning techniques. In machine-learning literature, moving-average methods (Seasonal Autoregressive Integrated Moving Average (SARIMA) model and Autoregressive (AR) model), Lazy methods (Instance-based-k (IBk) and Locally Weighted Learning (LWL)) and information-gain methods (REPTree and M5P) have been proposed. However, a comparison of these methods for real-world slope-movements has not been explored. The primary objective of this paper is to compare SARIMA, AR, LWL, IBk, REPTree and M5P methods in their ability to predict soil-movements recorded at the Tangni landslide in Chamoli, India. Time-series data about slope-movements from five-sensors placed on the Tangni landslide hill were collected daily over a 78-week period from July 2012 to July 2014. Different model parameters were calibrated to the training data (first 62-weeks) and then made to forecast the test data (the last 16-weeks). Results revealed that the moving-average models (SARIMA and AR) performed better compared to the lazy and information-gain methods during both training and test. Specifically, the SARIMA model possessed the smallest error compared to other models in test data. We discuss the implications of using moving-average methods in predicting slope-movements at real-world landslide locations.