An Improved Artificial Bee Colony Algorithm And Its Application In K-means Clustering

As one of novel swarm intelligence optimization algorithms, artificial bee colony algorithm has been developed rapidly in the last decade. The principle of artificial bee colony algorithm is inspired by the honeybees cooperative intelligent behavior, researchers abstracted the artificial bee colony algorithm from the whole process of gathering honey and applied it to solving the practical problems in our life. Due to the advantages of easy implemention, high search accuracy, strong robustness and better solution quality compared to classical optimization algorithms, artificial bee colony algorithm has attracted many scholars’ wide concern soon after it was proposed by the Turkey scholar Karaboga in 2005. According to the current references and papers, artificial bee colony algorithm has been applied to many fields and made good achievements, such as the traveling salesman problem, the artificial neural network training, the combinatorial optimization, the computer system optimization, the system and engineering design, the deployment of nodes in wireless sensor networks and the scheduling problems, and now researchers are exploring new application scenarios constantly. As a novel algorithm which is still in its infancy, the model of artificial bee colony algorithm remains to be perfected, many shortcomings may appear in its application process. For instance, while facing complex optimization problems, artificial bee colony algorithm is easy to fall into the premature convergence problem and the local optimal. Therefore, researchers started to explore how to improve the artificial bee colony algorithm theoretically and expand the application scenarios. The main research contents of this paper are as follows:Firstly, from the aspect of the overall optimization of algorithm performance, to cover the shortages of artificial bee colony algorithm,there are two points to be improved:First, at the stage of swarm initialization, this paper adopts a strategy with opposition-based learning to improve initial solution. According to the probability theory, the opposite solution has 50 percent chance of being close to the optimum solution comparing to the initial solution produced randomly. Therefore in order to increase population diversity and convergence rate, the algorithm selects the better one as initial population.Second, it’s about improvement of nectar source update formula. The nectar source update formula of basic artificial bee colony algorithm adopts the strategy of searching a new nectar source randomly in the field of objective nectar sources, which is so blind that the convergence rate slows down. This paper draws from idea of differential evolution algorithm and merges the mutation strategy of differential evolution algorithm into artificial bee colony algorithm. The paper also uses adjacent nectar source as a guide while searching new nectar sources, introduces random dislocation crossover strategy to collect helpful information in other dimensions, in addition, it introduces adaptive adjustment strategy. After the improvement of algorithm, the search of nectar source update formula is more motivated, the convergence rate of this algorithm increases, what’s more, the introduced adaptive dynamic adjustment variable balances better development capacity and search capacity of the algorithm. To prove the improved algorithm better, under premise of the same parameter setting, in the paper Ⅰ optimize a group of standard test functions respectively with the improved algorithm, the basic algorithm and other representative improved algorithms and compare to analyze the experimental results. The results indicate that the improved artificial bee colony algorithm has a higher optimization precision of functions.Secondly, applying the improved artificial bee colony algorithm (DEF-ABC) to the optimization of K-means clustering algorithm. As one of the usual algorithms in the process of clustering, K-means clustering algorithm is widely used because of its simple principle and easy implementation. In the process of searching and locating the clustering center, K-means clustering algorithm is very dependent on selection of initial clustering center, which causes the shortcoming that the algorithm is easy to fall into local optimal. In this paper, I optimize the process of searching clustering center with the improved artificial bee colony algorithm (DEF-ABC) which is already proved to be superior to the basic one to expect better clustering effect. While designing the experiment, I select some groups of standard test data from UCI machine learning database, and compare with other usual optimization methods, test and analyze them.