基于营销大数据的电力客户多维度信用评价模型研究

Research on Multi-Dimensional Credit Evaluation Model of Electric Power Customers Based on Big Data of Marketing

随着电网客户量的迅速增长,如何对电力客户进行准确的信用评价成为了一个重要问题,构建一个可以准确预测客户信用的模型是电力营销部门需要解决的关键问题.本文通过对已知信用评价模型的研究,结合集成学习思想,构建了一种基于XGB算法的客户多维信用评价模型,该模型通过采用多维度的营销数据,并基于特征重要度方法进行特征选择,采用极限梯度提升方法以及树模型对客户信用进行模型构建,计算不同节点上不同的增益值来获取最佳的预测效果,从而构建一个准确、稳定的客户信用评价模型.在经过客户历史数据进行模拟分析后,得出了客户信用评价结果,并与目前主流的机器学习算法包括基于梯度下降算法与基于树的传统算法进行比较,结果表明,与Logistic回归和其他3种基于树的模型相比,XGB模型不论是特征选择的准确性还是其分类性能都具有明显的优势.

At present,with the rapid growth of electricity customers,how to make an accurate credit evaluation for customers has become an important issue.Building a model that can accurately predict the credit of customer is a key step for the power marketing department.In this paper,we studied the credit evaluation model proposed by previous authors and combined the integrated learning idea to build a multidimensional credit evaluation model for customers based on XGB algorithm.Based on feature importance method,feature selection was carried out with the model by using multidimensional marketing data.The model for customer credit was built using gradient boosting method and tree model.By calculating the different gain values on different nodes to obtain the best prediction effect,an accurate and stable customer credit evaluation model was constructed.After simulation analysis of customer's historical data,the user credit evaluation results were obtained,which were compared with the current mainstream machine learning algorithms including gradient descent-based algorithms and traditional tree-based algorithms.The experimental study showed that the XGB model has significant advantages over Logistic regression and three other tree-based models,both in the accuracy of feature selection and classification performance.