Study On Application Of Artificial Bee Colony Algorithm

Most problems in many fields including computer science, management science, economics and control engineering can be attributed to solving a optimization problem. Meta-heuristic swarm intelligence algorithms are optimization methods that mimic natural biological group behaviour. Swarm intelligence does not require special information of the problems with obvious optimization effect, so become a new ways to solve the global optimization problems, therefore has become a research focus in the academics for a long time.The Artificial Bee Colony (ABC) algorithm is a new bi-inspired swarm intelligence optimization technique. It mimics the foraging behaviour of honey bees. One of the main features of the ABC is the division of labor. The swarms consist of three types of bees those are emplyed bee, onlooker bee and scouts. Individual of them are very simple but they can reflect high collective intelligence throught cooperating with each other in the group to find the food source, namely nectars. And the algorithm does exploit and explore in each iterative procedure. Those make it a simple, robust and efficient algorithm. Recently the ABC algorithm has been widely studied on and applied to many fields. The original ABC clustering algorithm suffers from slow to converge in early stage and easily to be trapped in local optima in the late stage. Firstly, this paper applies the wavelet mutation to the ABC clustering algorithm to overcome those problems. Experimental results on selected data sets show that modified the ABC is effective and outperforms the original ABC for clustering.Clustering analysis is a basic and important technique used in many fields such as data mining, machine learning, pattern recognition, image analysis, information retrieval, and bioinformatics. Traditional clustering methods are based on local greedy search so that they are sensitive to initial centers and are easily to trapped in local optima. This paper applies the discrete artificial bee colony algorithm to data clustering problem. Experiments results show that the DABC is also effective for clustering.Last decades semi-supervised learning becomes a hot topic in literatures. In real life there exist more data without labels, and it will cost more to give the data class labels in many scenarios. But getting constraints between data by uses or experts are relatively feasible. Thus clustering with pairwise constraints attracts more and more researchers to study the algorithms and apply it to many domains. The traditional clustering with instance-level constraints are sensitive to the assignment order, may not lead to convergence when over-constrained, and often trapped in local optima. This paper also explores the ABC algorithm to clustering with instance-level constraints. Experiments on tested data sets show that it will improve the accuracies of clustering significantly. It performs better when constraints sets are relative small.