Association Rule based Classifier Built via Direct Enumeration, Online Pruning and Genetic Algorithm based Rule Decimation

J.-M. Adamo (France)


Supervised learning, classifiers, direct association-rule mining, set covering problem, genetic algorithms.


Direct rule generation is a possible alternative to tree building for classifiers. Here, we use the association rule framework to build classifiers. The rule generator performs direct enumeration (no generation of candidate sequences or so, and no preliminary enumeration of large sets) with online pruning to keep combinatorial explosion under control. The rule set thus generated is ultimately and drastically decimated so that a final non redundant rule system with reduced learning bias is produced. Decimation is modeled as a minimum cost and minimal set covering problem solved with a genetic algorithm. Experiment results are presented and compared to results obtained with a tree building based classifier (C4.5).

Important Links:

Go Back