Data Mining and Hypothesis Refinement Using a Multi-Tiered Genetic Algorithm

Abstract

This paper details a novel data mining technique that combines set objects with an enhanced genetic algorithm. By performing direct manipulation of sets, the encoding process used in genetic algorithms can be eliminated. The sets are used, manipulated, mutated, and combined, until a solution is reached. The contributions of this paper are two-fold: the development of a multi-tiered genetic algorithm technique, and its ability to perform not only data mining but also hypothesis refinement. The multi-tiered genetic algorithm is not only a closer approximation to genetics in the natural world, but also a method for combining the two main approaches for genetic algorithms in data mining, namely, the Pittsburg and Michigan approaches. These approaches were combined, and implemented. The experimental results showed that the developed system can be a successful data mining tool. More important, testing the hypothesis refinement capability of this approach illustrated that it could take a data model generated by some other technique and improves upon the overall performance of the data model.