Study On The Improved Artificial Bee Colony Algorithm Based Fuzzy Clustering

With the fast development of Internet related technologies, the amount of data is growing exponentially. It has become one of the urgent problems that how to get information with social value and business value from these massive data. Clustering analysis is a kind of techniques that finds the unknown characteristics and knowledge from the data through the appropriate clustering algorithm in the absence of prior knowledge.The fuzzy clustering algorithms describe the relationship between patterns more conform to the reality; therefore it becomes one of the important research focuses. The fuzzy c-means (FCM) algorithm is a commonly used algorithm which based on the objective function. It describes the clustering problem as a mathematical function optimization problem with constraints, and obtains fuzzy clustering of data objects by solving the mathematical problem. The FCM algorithm has advantages of simple model, easy programming, low complexity, good clustering effect, etc. It was successfully applied to machine vision, image processing and other fields. However the clustering results are over dependent on the initial cluster center, and are easily falling into local minimum.The artificial bee colony optimization algorithm simulates the foraging process of honey bees in the nature. It has advantages of parallelism, less parameters, strong robustness, etc. Aiming at the disadvantages of premature convergence, low precision and the slow convergence speed in the late, it is proposed an artificial bee colony algorithm based on improved search strategy and chaos mechanism (ABC-SC). In this algorithm, in order to balance exploration and exploitation abilities, the search direction is positively guided by the historical average optimal solution. In order to ensure the population diversity, chaotic mechanism is used in individual mutation on several dimensions if the population falls into local extremum. It is made the test experiment that uses ABC-SC algorithm to optimize the six benchmark functions. The results show that the ABC-SC algorithm has higher precision and faster speed than the basic ABC algorithm and other improved ABC algorithms.Aiming at the existing problems of the fuzzy c-means clustering, the three fuzzy parameters that the objective function, the fuzzy membership degree and cluster center were adjusted to avoiding falling into local minima. Use ABC-SC algorithm to optimize the initial cluster center, and then use the output of ABC-SC algorithm as the initial cluster center of FCM algorithm. This method overcomes the shortcoming that it is sensitive to the initial value. Clustering experiments based on UCI and artificial data sets show that this new algorithm not only overcomes the disadvantages of FCM algorithm, but also has higher clustering accuracy and better performance.