Rule induction using a reverse Polish representation
Guy Davenport, Mark Dermot Ryan, Victor J. Rayward-Smith · 1999
It is often necessary to extract simple and understandable rules from databases containing inconsistent records and/or irrelevant fields. In this study we have assessed the feasibility of using a genetic programming (GP) approach to extract a single rule to describe such data. Instead of a tree structure we use a Reverse Polish (post-fix) representation. To assess the performance of the GP algorithm, it is compared to a steepest ascent hill climber algorithm and C5.0, a commercially available data mining algorithm (Quinlan, 1997). On the datasets used, the GP algorithm out-performs both C5.0 and a steepest ascent hill climber in the simplicity and, in most cases, the accuracy of the expressions produced.