Report for PAKDD 2006 Data Mining Competition

The task requires that a classifier model can give high prediction accuracy over 3G customer, while keeping reasonable misclassification rate for 2G customer. In this report, we firstly learn a classification model with high AUC(The area under the ROC curve) based on our latest research result in classification algorithm, and then find a optimal decision threshold from the resulting ROC(Received Operating Characteristic) curve plot. At last, the human understandable knowledge learned from this real-world problem are also presented. 1 Understanding of the problem The PAKDD 2006 Data Mining Competition offered a excellent opportunity for the application of data mining about identifying potential telecom customer. In this competition, the task is to learn a classification model that can predict potential 3G customer as much as possible. The original training data set consisted of 20,000 2G network customers and 4,000 3G network customers, which has been provided with more than 200 data fields. A 3G customer is defined as a customer who has a 3G Subscriber Identity Module (SIM) card and is currently using a 3G network compatible mobile phone. Based on the above problem description, we may need to deal with the following three challenging problems. First, a 2G customer may transfer to 3G customer at any time, or a 3G customer may be a 2G customer before. Thus, it is not surprised that we may encounter the same customer with different class label. Second, rather than giving a properly defined misclassification cost function, a subtle true positive rate, identifying 3G customer as much as possible, is needed. Since for any classifier, there is always trade off between True Positive rate and True Negative rate, where true class refers to 3G customer and negative class refers to 3G customer. Thus, the underlying requirement is to maximize the True Positive rate, while keeping the reasonable the True Negative rate. Third, human understandable knowledge should be presented, which are useful for making business decision, such as different promotion strategies for vary customer groups.