Improving Data Segregation with a Graph Based Evolutionary Algorithm
Submitted to Annie 2002

Daniel Ashlock, Kenneth M. Bryden, Peter Johnson, and Douglas McCorkle

Abstract PDF eprint

Interpolation, compression, or even use of a large data set is enhanced if the data set can be partitioned into subsets with a more uniform internal character. This paper presents a technique for more rapid automatic segregation of data sets with an evolutionary algorithm. We improve performance of our evolutionary algorithm by imposing a graphical geography that, we conjecture, slows the spread of information within the evolving population and so retards premature convergence. We present results on a trial data segregation problem for 23 different graphical geographies. Change of geography has a statistically significant impact on performance.