KReSIT Logo Kanwal Rekhi School of Information Technology
IITB Logo
IIT Bombay
 
 Research
 Groups
 Publications
 Projects
 Student Projects
 Ongoing Projects
 Past Projects
 Proposed Projects-2005
 Projects Archives
 ASL BE/MCA Projects
 Sponsored Projects
 Seminars
 Labs
 Techtalks
 

Home > Research > Projects > Students > Previous 

Students' Projects


Using Multitude of Time Series Forecasting Models to Improve Forecast

Abhishek Seth

Abstract


We study the effect of combining a large number of forecasting models(called experts) on the performance of forecasting. The present work is carried out on using dynamic combination of experts by determining a good combination of experts at each point of time, considering only the immediate window of past forecasts. We have proposed some kinds of windows that can be used by combining model, and demostrated the working of some of them. We have studied the effects of various windows on the performance of dynamic combination, and made some attempts to find out any optimal window if there exists one. Due to the increase in the number of experts dramatically, we have discovered some nontrivial phenomena like Outlierness and Role Reversal and discussed some preliminary ways to handle them. We tested some of the techniques (on the shelf and newly developed) of combining a large number of experts, which are found working good.








Printer friendly    Comment
  Copyright © 2004 KReSIT, IIT Bombay. All rights reserved sitemap    
  Kanwal Rekhi School of Information Technology, Indian Institute of Technology Bombay, Powai, Mumbai - 400 076.
+91-22-2576 7901/02. Fax: +91-22-2572 0022
Designed by Kamlesh