Abstract:
An approach to fuzzy rule induction inspired by the foraging behaviour of ants is presented. The implemented system - FRANTIC - is tested on a real classification problem against two other fuzzy rule induction algorithms, one with an emphasis on rule comprehensibility, and the other on rule accuracy. The results obtained highlight FRANTIC’s ability to balance the tradeoff often encountered between predictive accuracy on the one hand, and ruleset comprehensibility on the other. FRANTIC’s actual and potential strength when applied to realworld large datasets is highlighted, while its limitations and the possible ways of overcoming them are also discussed.
