L-RBF: A Customer Churn Prediction Model Based on Lasso + RBF

With the development of market economic, customer churn prediction play a critical role in the company management. How-ever, customer information is complex, it contains multi-dimensional features. What's worse, the number of customer churn very small among the whole consumers, and the features of customer are dynamically changing, which is challenging for traditional statistical methods. Fortunately, as the rise of edge computing, more computing related to data-intensive applications will be de-centralized to edge smart terminals. Moreover, edge computing focuses on real-time, short-cycle analysis, which is useful for dealing with dynamic changes problems in customer features. Therefore, to address these issues about customer messages, this paper proposes a simple and effective model to predictive customer churn with a higher accuracy, named L-RBF. The basic idea is to utilize Lasso Regression algorithm (Lasso) for optimizing Radial Basis Function Neural Network (RBF). At first, placing customer features on edge terminals, and Lasso is utilized to extract the correlated features of customer churn. Then, according to the correlated features and lasso regression equation, we can automatically set the topology and parameter information of RBF. Finally, experimental results indicate that L-RBF has a higher recall rate and stronger prediction classification ability compared with the previous work, included Logistic Regression (Log-R), RBF and Boosting.