Prediction of tariff package model using ROF-LGB algorithm

With the slowing growth of the telecommunication market and the intense competition for existing customers, Customer Churn Management has become a crucial task for all mobile network operators. Recommendation models based on customer behaviors are widely used by operators to provide diverse telecom tariff packages for suitable people and thus improve customer satisfaction. To address the low precision rate and data granularity of prior studies, this study combined rotation forest (ROF) and LightGBM and construct a hybrid algorithm (ROF-LGB). Grid search method was used in parameter tuning, and ten-fold cross-validation method was used to prevent overfitting. Using mobile data generated by operators, ROF-LGB method was tested and compared with other five traditional machine learning methods. The results showed that ROF-LGB method achieved better performance with better precision rate and execution efficiency in telecom tariff package recommendation.