The Global Clustering Algorithm Based On AFS Neighborhood

As we all know, cluster analysis has wide and deep applications in many fields, and has a very important significance to the guidance to people’s decision-making and study. K-means algorithm is a basic division of the cluster analysis approach and it has been one of the most popular cluster methods, for it’s popular and easy to understand, and also for its high computational efficiency. Therefore, its application and improvement has got widely attention.The AFS neighborhood is a characteristic of AFS theory describing the individual or object in the collection of“position”, which is of advantage handling multiple types of data. The paper attempts to make the AFS neighborhood concept applied to K-means clustering algorithm, and achieve the purpose of dealing with varieties of data types. Firstly, the AFS neighborhood-based global clustering algorithm is proposed based on the introduction of the K-means clustering algorithm and AFS neighborhood knowledge. The algorithm gets improved based on the overall K-means clustering algorithm and proposes a new look for the next initial class cluster center; on the other hand, with the knowledge of AFS theory, it makes reduction to the properties of the data set, and then with the AFS topology and o-domain, it produces a relative distance between the dataset object and the relative distance from the cluster center used in the initial class identification and class cluster center update so as to get the final clustering results. Finally, the paper carries out the clustering experiments to Iris, Wine, and etc. ---even groups of numerical date, non-numeric data Balloon data of the machine learning database and randomly generated artificial data sets with noise point. The experiment results show that the algorithm has better clustering results, the processing power of multiple data types and noise immunity, and then we achieve the purpose of the experiment.Meanwhile, in order to truly make the theory applied to practical, the clustering algorithm is developed into an application running under Windows according to Matlab GUI technology, and then the Iris data used as a test case provides a reference to its practical application.