Graduation Semester and Year

2014

Language

English

Document Type

Thesis

Degree Name

Master of Science in Computer Science

Department

Computer Science and Engineering

First Advisor

Ramez Elmasri

Abstract

Spatio-temporal trajectories are time series data that represent movement of an object over the time. Hidden Markov Models (HMM), a variant of Markov Models (MM), were first applied at a large scale to speech recognition but have also been used in time series prediction by analyzing trends in historical time series data. In this research, we propose a storm prediction model using a HMM built from overall storm trajectories derived from raw rainfall data. This HMM is built by assuming the states are associated with clusters created by clustering the locations of each storm from the overall storm trajectories. Then we learn the variation of transitional information and spatial information present in the given set of trajectories using Baum-Welch algorithm. This learning is performed by building a HMM that for each cluster contains multiple state instances that represent this cluster and can learn to reflect variations in the information within a cluster. Results from experiments showed that the prediction gets better when the number of state instances representing each cluster increases. For example, the average distance value between actual location and location predicted by a model with 5 clusters and 5 state instances per cluster is approximately 15% smaller than the average distance value between actual location and location predicted by a model with 5 clusters and 3 state instances per cluster, which means that the predicted location gets closer to the actual location with more state instances. It was also found that the prediction gets better when the number of clusters increases. Apart from introducing a new prediction model for this type of data, we also propose a modified algorithm that creates overall storm trajectories much faster than existing algorithm.

Disciplines

Computer Sciences | Physical Sciences and Mathematics

Comments

Degree granted by The University of Texas at Arlington

Download

Share

COinS