混合Markov与Bayes的客户欠费预测模型

为有效规避客户欠逃费和实现差异化处置,针对具有长期定时付费特征的后付费类服务,提出混合马尔科夫与贝叶斯的客户欠费预测模型,基于全客户多要素信息增益分析,给出潜在欠费客户的欠费概率,为客户欠费预警和处置提供全面、客观、精细的决策信息,并支持客户差异化处置.首先,基于其付费特点,建立k序马尔科夫模型,计算客户的初始欠费概率;然后,融合客户基本属性、行为特征和欠费信息等要素,基于条件互信息和爬山法生成目标贝叶斯网络,对初始欠费概率予以修正,形成客户最终欠费概率;最后,基于实际数据进行实证分析,验证了该模型的有效性.

A prediction model of customer arrears based on hybrid Markov and Bayesian

In order to analyze the post-paid services with the characteristics of long term and on time pay,we put forward a prediction model based on hybrid Markov and Bayesian. It is based on the multi-factor information gain of all the customers,and computes the potential owe customers' probability of arrears.Moreover,it can provide comprehensive,objective and subtle decision information to the warning of customer arrears and disposal,and it can support differentiation treatment.First of all,we build the k-order Markov model based on the characteristics of the pay,then calculate the customers' initial probability.Secondly,we merge the customers' basic attributes,behavior feature and own information.Then,using the conditional mutual information and the hill climbing algorithm to generate the target Bayesian network to modify the initial probability of arrears,which form the final client own probability.Finally,through experiment by using the real data,we prove that this predict model is efficient in customer prediction.