Research On An Improved Clustering Algorithm Of K_means

With the rapid development of network and the steady progress of information technology, a variety of data expand rapidly. How to find useful information and to classify them has been becoming increasingly urgent. The appearance of data mining technologies makes it possible to process large amounts of data. Clustering is one of important data mining technology and has been becoming a research hotspot.In this paper, two improving points of the k_means clustering algorithms are proposed to solve the limitation of its sensitivity to outliers and the selection of initial points. Firstly, after outlier detection problem of the k_means algorithm is studied deeply, a grid-based data pre-processing algorithm is proposed. This method detects outliers by dividing the dataset grid. Secondly, on the base of analyzing the selection of initial points, an algorithm of initial point selection is proposed. This method directly chooses the initial point based on the grid-base data's pre-processing, so that the initial point is more reasonable and closer to the actual cluster center. Finally, during the process of k_means clustering, those two algorithms are used to deal the outlier and to select the initial point and the improved k_means algorithm is given.The improved k_means algorithm is verified and analyzed by experiments. Experimental results show that the improved k_means algorithm improves the accuracy of clustering in a certain extent.