Indian Journal of Science and Technology

Abstract

Objective: The paper aims in presenting a prediction model by using Support Vector Machine (SVM) technique which is meant to possess a strong capability to predict landslides by forecasting rainfall dataset using BigData concept. Methods: The dataset has been taken for the Cherapunjee region which receives the highest intensity of rainfall in India. The aim is to predict the landslide occurrence and classify the risk level associated with the landslide. To improve the reliability in landslide prediction, the proposed model uses pre-processing for removing null values in the dataset. After getting the pre-processed dataset, it will apply normalization, then SVM training and finally the Testing process. Thus the Support Vector Machine concept proved to exhibit a large degree of flexibility in handling tasks of varied complexities because of the non-linear boundary functions. Findings: The study concludes that SVM proved to be an efficient technique to forecast the landslides by predicting the rainfall in advance. The comparative results of SVM in regard with Artificial Neural Networks were proven. The study has been done specifically for Cherrapunjee region and can be implemented for any landslide prone area. Novelty/Improvement: Researchers worldwide are having a great pace to develop early prediction mechanisms for natural hazards. The study uses Radial Basis Function as an initial parameter for predicting the risk level classification of landslide. The novelty is in providing an initial selection of the kernel parameter in order to save the time on finding the best parameters
Keywords: BigData, Hadoop, Rainfall Data, SVM