Back to the program.
Building Rule-Based Classifiers with Ant Colony Optimization
David Martens
Department of Decision Sciences and Information Management


AntMiner+ is a classification technique based on ant colony optimization, and builds comprehensible rule-based classifiers. The key differences between the proposed AntMiner+ and previous ant-based classification techniques are the usage of the better performing MAX-MIN ant system, a clearly defined environment for the ants to walk through, and the ability to include interval rules in the rule set. The commonly encountered problem in ant systems of setting system parameters is dealt with in an automated, dynamic manner. We applied AntMiner+ to a variety of data sets and compared its performance with several state-of-the-art classification techniques. Our benchmarking experiments indicate competitive results and show that when considering both accuracy and comprehensibility AntMiner+ ranks at the top. The importance and possibility of including prior domain knowledge in AntMiner+ is also briefly discussed.


data mining, classification, AntMiner+, MAX-MIN ant system


  1. De Backer M., Haesen R., Martens D., Baesens B.. (2005) A stigmergy based approach to data mining. In Proceedings of the 18th Australian Joint Conference. Lecture Notes in Computer Science, Sydney, Australia. pp. 975 - 978.
  2. Martens D., De Backer M., Haesen R., Baesens B... A MAX-MIN Ant System Working Towards Comprehensible Classifiers. Under Review IEEE TEC.