The Application Research Of Fuzzy Clustering Algorithm In Mobile Customer Churn Prediction

With the rapid development of the economy and the telecommunications industry, the mobile customers churn problem exists further. Therefore, predicting customer churn has become a key measure for corporation to maintain economic efficiency and the development of enterprises. In recent years, due to the powerful advantage of data analysis, data mining technology has been widely applied in customer churn prediction. At the same time, as the fuzzy theory becomes matured, fuzzy clustering algorithm also begins to be widely used in a variety of practical problems.FCM algorithm is the most widely used fuzzy clustering algorithm. The paper studied FCM algorithm in detail. In order to improve the shortcoming that FCM algorithm is easy to fall into local optimal solution, the paper proposed a FCM algorithm based on optimization genetic algorithm(IGC-FCM).First, the paper presents a real-coded strategy for genetic algorithm to prevent the shortcoming that encoding length may be too long and the algorithm may precocious, and then the paper proposes an adaptive mutation operator, which can expand the global search ability of the genetic algorithm, making genetic algorithm have better fitness; then, the paper combines the optimization genetic algorithm with traditional FCM algorithm(IGC-FCM), and the new algorithm not only retains the advantages of fast convergence of the traditional FCM algorithm, but also it can expand the global search ability of the algorithm and it can suppress traditional FCM local convergence.Then, we select the mobile business data to create customer churn prediction model based on improved FCM algorithm in this paper. Due to mobile services data is large and integrated complex, we do a lot of pre-processing of the data before modeling, including data cleaning, data integration and transformation, data statute, finally we extract some customer data as the dataset of the predictive model. Then we establish a churn prediction model with IGC-FCM based on the WEKA platform, and then classify the customers according to the customer segmentation matrix proposed by Modisette, then the paper analysis each generated clustering, the paper also estimates the accuracy and coverage performance of the prediction model. Finally, the paper summarizes the theoretical research and experiments, and also describes the future research directions.