KES'10 Proceedings of the 14th international conference on Knowledge-based and intelligent information and engineering systems: Part I
| Hi-index | 0.00 |
As markets become increasingly saturated, churn prediction and management has become of great concern to many industries. A company wishing to retain its customers needs to be able to predict those who are likely to churn and will make those customers the focus of customer retention efforts. Today Customer data has properties of large samples, high dimensions and more noises. In response to the limitations of existing feature selection in churn-prediction, we introduce and experimentally evaluate Support vector machine-recursive feature elimination attribute selection algorithm. It can identify key attributes of customer churn, rule out the related and redundant attributes, and reduce the dimensions of data. It is more important that this algorithm is related with the followed classification learning algorithm, so it can be better integrated in churn prediction. The empirical evaluation results suggest that the proposed feature selection algorithm extracts less key attributes and exhibits better satisfactory predictive effectiveness than other three comparable attribute selection algorithms.