Research And Application Of Clustering Ensemble Algorithm In Customer Segmentation

With the deepening development of information technology, information growth rapidly, meeting the demand of information to people it has also brought new challenges, how to find valuable information in these large and complex flood of data has become increasingly important. The emergence of data mining brought new opportunity to deal with these problems, so that it can be applied to different industries quickly. Customer segmentation is a very good application scenarios of data mining, against the insufficient for traditional customer segmentation techniques, applying data mining techniques to customer segmentation can get better segmentation results.Cluster analysis is one of the common method of data mining, has its unique advantages in the application of customer segmentation, and clustering ensemble is integrated with multiple features of the cluster classification to get a more stable and accurate classification. Firstly, this thesis optimized the selection of cluster centers of the K-means clustering algorithm, in connection with the multiple maximum density parameter value to obtain corresponding samples points as the initial cluster centers, appearing time-consuming, poor accuracy defects, proposed a scheme based on second density improvements. When existing multiple maximum density parameter values, obtaining the second density value through these maximum density parameter values corresponding to the sample points, and then based on the parameter values to obtain the optimal initial cluster centers, this program has shorten the time to select the initial cluster centers, while increasing the accuracy of clustering; Bring a model structure into the iterative process of divided clusters, it used to store cluster label sample points and the distance to the cluster center, compared with the next iteration of the results, adjusting the computing model through the results so that it can shorten time and reduce iterations computation purposes; and then verify the effectiveness of the algorithm by experiments.According to the clustering ensemble theory to integrate multiple single clustering result in this paper. Firstly, by repeating the sampling method to generate a subset of data and divided by the improved clustering algorithm, generating a plurality of clusters based devices; calculating the difference of each base clustering according to the Normalized Mutual information (NMI), and then choose the clusters based devices according to the differences, voting the clusters based devices after selection by Majority voting method and get the final results.After the detailed analysis of the traditional segmentation technology and customer segmentation theory based on data mining, used the previous clustering ensemble algorithm for customer segmentation of communication industry. Processing the customer data firstly, then segments these processed data by algorithms, extracting characteristics of different customer groups based on the group’s segment results, and then make the interrelated plans of marketing strategies, Experimental results show that work in this paper has practical significance to the application.