Research On The Mobile Customer Churn Alarming Model Of China Mobile

With the reform and development of the communications industry,fierce competition between communication operators has gradually appeared.In parallel with the rapid development of the Internet,more and more new applications being developed.Traditional voice calls,SMS,MMS,etc,are gradually replaced by applications such as Wechat,QQ.Since the communications markets have been squeezed,the competition becomes more fiercer.Recruiting new customer and keeping old customer have become an important business philosophy.And reducing the loss rate of customers effectively is the key to enhance the competitiveness for the operators.Based on China Mobile of Beijing Company communications market,the paper selected 65 original indicator and 30 derived indicators as initial indicators set.And 20 important indicators were chosen from the initial indicators set by attribute reduction algorithm of rough set theory,aiming at reducing redundant information between the variables.After identifying the 20 indicators involved in model,the paper chose C5.0 algorithm of decision tree to build customer churn alarming models to portraits the features of lost customers,and put forth effective marketing suggestions combined the conclusions of the model.In the paper,validating the model with two different validation set.Meanwhile,accessing the decision tree model with 5 derived indicators of confusion matrix.Accuracy rate,classification error rate,false alarming rate,hit rate and coverage rate performed outstandingly.The customer churn alarming models passed the validation and the assessment.According alarming rules of the decision tree model,the main conclusions of the paper are as followed:(1)online months is the most significant segmentation indicator.Among the customers whose online months are less than 41 involved in the model,98.9% of them were lost.(2)Whether the customers owes fee is another significantly feature.(3)The demand of GPRS flow,monthly recharge amounts,whether downtime,and whether hard bundled are benefit for forecasting the churn customers.Based on the conclusions,several marketing advices were given as followed:(1)Writing SQL in the database to extract the data corresponding to the potential churn customers of every group.So that the features of every group can be identified(2)Providing scores feedback activity for those old customer.While keeping junior customers,price concessions can be offered.(3)Developing attractive retention programs for the young around 20,such as offering free flow or recharge for disk.(4)Filtering out those customers those credit level is not high at present but with good credit records.And providing them free trial of services which those high credit level customers can enjoy.