Research On Association Rule Mining Algorithm Application In Customer Churn Prediction

With the comprehensive adjustment of the telecommunication enterprises in China, the competition in Telecommunication industry is more severe. The competition leads to the customer churning, which becomes one of the important elements to reduce the profit of the telecommunication enterprise. Then using data mining to scientifically analyze the characteristics of the churned customer whereby to predict the churn tendency of the existing customer has become an important research in Telecommunication industry. And association rule mining technology based on its so many advantages, turns into one of the most prospective data mining technology in customer churn prediction, and receives more and more attention from the researcher.This dissertation analyzes the problems in existing research via the literature review of the customer churn prediction in telecommunication industry and association rule algorithm, and emphasizes that the existing association rule algorithms can not mining the rules of the few churned customers with big value, and the efficiency is low. Accordingly, this dissertation brings forward an improved multidimensional association rule algorithm based on attribute reduction and probability weighted item, then the approach is experimentally evaluated, which suggest that approach is fairly effective. The contributions of this dissertation are as follows:Firstly, by reviewing some of research literature related to customer churn prediction and association rule mining, analyzes the classification and value evaluation of existing algorithm; compares the advantages and disadvantages of the existing multidimensional association rule approach in customer churn prediction; points out that the telecommunication industry data is redundancy and those algorithms are inefficient ;moreover, did not attach importance to few churned customers with big value, which lead to the inefficiency in customer churn prediction.Secondly, according to the previous analysis, a novel multidimensional association rule algorithm based on attribute reduction and probability weighted item is put forward. The main idea is as follows: Firstly, do attribute reduction with the data of big value customers to form a table, subsequently, introduce the conception, definition and formula of the improved algorithm to mine the rules of the few churned customers with big value, and use"lift"to evaluate the rules in order to find strong association rules of the customer churning.Thirdly, the experiment conducted by Powerbuilder6.5 programming indicated that new algorithm is more efficient in customer churn prediction and can mine rules of few churned customer with big value which traditional algorithm can not mine. The new algorithm shows better predictive results and can make the telecommunication enterprises retain big value customers more pertinently.