Exploring different rule quality evaluation functions in ACO-based classification algorithms

Abstract

The μAnt-Miner algorithm is an extension of the well-known Ant-Miner classification rule discovery algorithm. μAnt-Miner utilizes multiple pheromone types, one for each permitted rule class. An ant would first select the rule class and then deposit the corresponding type of pheromone. In this paper, we explore the use of different rule quality evaluation functions for rule quality assessment prior to pheromone update. The aim of this investigation is to discover how the use of different evaluation function affects the output model in terms of predictive accuracy and model size. In our experimental results, we use 10 different rule quality evaluation functions on 13 benchmark datasets, and identify a Pareto frontier of 4 evaluation functions.