One aspect of evolutionary computing as a method of data mining, is its intrinsic ability to drive model selection according to a mixed set of criteria. Based on natural selection, evolutionary computing utilizes evaluation of candidate solutions according to a ftness criteria that ~xfight or might not share the exact same implementation as the metric used to measure the performance of the selected solution. This paper presents the results of using four different fitness functions to evolve nai’ve Bayesian networks based on a combination of Mean Absolute Percent Error and Worst Absolute Percent Error values tbr individual population members. In addition to the error measurements tiom both the training and lbrecast evaluations, data is presented that shows APE lbr individual members during the tbrecast generation and evahtation phase.
[1]
Wen Lea Pearn,et al.
Assessing the statistical characteristics of the mean absolute error or forecasting
,
1991
.
[2]
Farhad Kamangar,et al.
Inferencing Bayesian Networks from Time Series Data Using Natural Selection
,
2000,
FLAIRS.
[3]
John H. Holland,et al.
Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology, Control, and Artificial Intelligence
,
1992
.
[4]
Farhad Kamangar,et al.
Forecasting of time series data using naturally selected bayesian networks
,
2000
.
[5]
David E. Goldberg,et al.
Genetic Algorithms in Search Optimization and Machine Learning
,
1988
.