Genetic Programming as a Data-Mining Tool

Abstract

Genetic Programming (GP) has increasingly been used as a data-mining tool. For example, it has successfully been used for decision tree induction (Marmelstein and Lamont, 1998; Choenni, 1999), data fusion (Langdon, 2001) and has also been used for the closely related problem of intelligent text retrieval on the Internet (Bergstrom, Jaksetic and Nordin, 2000). Indeed its ability to induce a program from data makes it a very promising tool for data mining applications. It has been successfully applied in many different fields and has even produced results that have exceeded those produced by other means. For example it has been used to evolve chemical structures (Nachbar, 2000) using a quantitative structure activity relationship model. It has also had success in spacecraft attitude control (Howley, 1996) where near-minimum spacecraft attitude manoeuvres were evolved which outperformed previous hand-coded solutions. It has also been used in quantum computing, where it was used to evolve quantum algorithms (Barnum, Bernstein and Spector, 2000). In this work, it rediscovered known algorithms such as Deutsch’s Early Promise Problem and discovered quantum results that experts did not think could exist, for example, AND-OR query problem. These examples demonstrate the versatility and potential of GP.