Type of Document	Thesis
Author	Lay, Nathan Stephen
Author's Email Address	nlay@fsu.edu
URN	etd-11082009-145713
Title	Supervised Aggregation of Classifiers using Artificial Prediction Markets
Degree	Master of Science
Department	Scientific Computing, Department of
Advisory Committee	Advisor Name Title Adrian Barbu Committee Chair Anke Meyer-Baese Committee Co-Chair Tomasz Plewa Committee Member
Keywords	Machine Learning Aggregation Random Forest
Date of Defense	2009-11-05
Availability	unrestricted
Abstract Prediction markets have been demonstrated to be accurate predictors of the outcomes of future events. They have been successfully used to predict the outcomes of sporting events, political elections and even business decisions. Their prediction accuracy has even outperformed the accuracy of other prediction methods such as polling. As an attempt to reproduce their predictive capability, a machine learning model of prediction markets is developed herein for classification. This model is a novel classifier aggregation technique that generalizes linear aggregation techniques. This prediction market aggregation technique is shown to outperform or match Random Forest on both artificial and real data sets. The notion of specialization is also developed and explored herein. This leads to a new kind of classifier referred to as a specialized classifier. These specialized classifiers are shown to improve the accuracy of prediction market aggregation even to perfection.
Files	Filename       Size       Approximate Download Time (Hours:Minutes:Seconds)     28.8 Modem   56K Modem   ISDN (64 Kb)   ISDN (128 Kb)   Higher-speed Access    Lay_N_Thesis_2009.pdf 432.81 Kb 00:02:00 00:01:01 00:00:54 00:00:27 00:00:02