| Summary: | When and why people change their mobile phones are important issues in mobile communications industry, because it will impact greatly on the marketing strategy and revenue estimation for both mobile operators and manufactures. It is a promising way to take use of big data to analyze and predict the phone changing event. In this paper, based on mobile user big data, first through statistical analysis, we find that three important probability distributions, i.e., power-law, log-normal, and geometric distribution, play an important role in the user behaviors. Second, the relationships between eight selected attributes and phone changing are built, for example, young people have greater intention to change their phones if they are using the phones belonging to the low occupancy phones or feature phones. Third, we verified the performance of four prediction models on phone changing event under three scenarios. Information gain ratio was used to implement attribute selection and then sampling method, cost-sensitive together with standard classifiers were used to solve imbalanced phone changing event. Experiment results show our proposed enhanced backpropagation neural network in the undersampling scenario can attain better prediction performance.
|
|---|
