Inferencing Bayesian Networks from Time Series Data Using Natural Selection

Andrew J. Novobilski and Farhad A. Kamangar, University of Texas at Arlington, USA

This paper describes a new framework for using natural selection to evolve Bayesian Networks for use in forecasting time series data. It extends current research by introducing a tree based representation of a candidate Bayesian Network that addresses the problem of model identification and training through the use of natural selection. The framework constructs a modified Naïve Bayesian classifier by searching for relationships within the data that will produce a model for the underlying process generating the time series data. Experimental results are presented that compare forecasts in the presence of multiple sources of information made using the naturally selected belief network versus a random walk.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.