No abstract

This paper proposes a novel tool known as Bee for Mining (B4M) for classification tasks, which enables the Bees Algorithm (BA) to discover rules automatically. In the proposed B4M, two parameters namely quality-weight and coverage-weight have been added to the BA to avoid any ambiguous situations during the prediction phase. The contributions of the proposed B4M algorithm are two-fold: the first novel contribution is in the field of swarm intelligence, using a new version of BA for automatic rule discovery, and the second novel contribution is the formulation of a weight metric based on quailty and coverage of the rules discovered from the dataset to carry out Meta-Pruning and making it suitable for any classification problem in the real world. The proposed algorithm was implemented and tested using five different datasets from University of California, at Irvine (UCI Machine Learning Repository) and was compared with other well-known classification algorithms. The results obtained using the proposed B4M show that it was capable of achieving better classification accuracy and at the same time reduce the number of rules in four out of five UCI datasets. Furthermore, the results show that it was not only effective and more robust, but also more efficient, making it at least as good as other methods such as C5.0, C4.5, Jrip and other evolutionary algorithms, and in some cases even better.