逻辑回归算法在通信GPRS业务中的应用

Application of Logistic Regression Algorithm in GPRS Business

为了对通信GPRS业务用户的流失情况进行有效的预测,对预测中常用三种算法的优劣势进行了比较,选取能很好处理0/1分类问题的逻辑回归算法,并基于逻辑回归算法建立了GPRS业务流失预测模型.提取了广东省移动公司GPRS业务用户流失概率最高的前5%和10%用户,通过查准率、查全率和提升率这三个指标对该模型进行检验,发现该模型定位流失用户的准确率和确定流失用户的覆盖率都是相当高的,说明其能对GPRS套餐使用客户的流失情况进行有效地预测.另外,通过把利用逻辑回归算法与利用决策树算法建立的预测模型的应用效果进行了对比,结果充分说明了利用逻辑回归算法建立的GPRS业务流失预测模型在实际应用中更具优越性.最后,根据该模型解在决实际预测问题中的效果,进一步验证了其具有很强的实用性.

In order to effectively predict the loss of the users of communication GPRS business, a comparison between three algorithms was made that are commonly used in practical predictions. The Logistic algorithm which is better for the 0/1 classification was selected,on the basis of which the model for predicting the loss of users of GPRS business was built. The top 5% and 10% users of the GPRS business in the Guangzhou Mobile Company were extracted whose loss probabilities were the highest. The model was tested with three indicators, including accuracy rate, recall rate and increase rate. It is found that both the accuracy rate of positioning the loss of the users and the coverage rate of confirming the loss of the users are fairly high,showing that the model can effectively predict the loss of the GPRS package users . In addition, the applications of the two predicting models built respectively on the Logistic algorithm and the Decision tree algorithm were compared. The comparison shows that the model built on the Logistic algorithm is superior in practical use. And its practicality was further verified.