Earthquake Forecasting by Parallel Support Vector Regression Using CUDA

Abstract

Earthquakes are a devastating natural hazard that can wipe out thousands of lives and cause economic loss to the geographical location. Seismic stations continuously monitor and gather data regarding the vibration and movement of the ground at a particular site. The collected data is processed by the model to forecast the occurrence of earthquakes in the Caribbean region. This paper presents a Parallel Support Vector Regression (PSVR) model to forecast earthquakes using Graphic Processing Unit (GPU). In the implementation of a PSVR using GPU, Computing Unified Device Architecture (CUDA) framework is utilized, which is a famous programming structure for General Purpose Computing on GPU. This newly computed PSVR model shows considerable improvement in training speed and achieved an accuracy of 92% when compared with Scikit Learn and LibSVM library on Central Processing Unit (CPU) and GPU.