Improved ant colony algorithms for data classification
Mohamed Hamlich, Mohammed Ramdani · 2012
In this paper we propose an extension of classification algorithm based on ant colony algorithms to obtain rules from data. Continuous attributes are handled by using the concepts of fuzzy logic. The ant colony algorithms transform continuous attributes into nominal attributes by creating clenched discrete intervals. This may lead to false predictions of the target attribute, especially if the attribute value history is close to the borders of discretization. Continuous attributes are discretized on the fly into fuzzy partitions that will be used to develop an algorithm called Fuzzy Ant-Miner. Fuzzy rules are generated by using the concept of fuzzy entropy and fuzzy fitness of a rule. Fuzzy Ant Miner algorithm is based upon the basic ideas published in 2010 [14] and 2011 [15]. The results obtained are very encouraging.