Research And Application Of Power Customer Complaint Prediction Model Based On Data Mining

With the continuous deepening of power market reforms,the people’s requirements for power supply capacity and service quality are also increasing.In the context of big data,the use of data mining technology to understand market needs and customer demands,and comprehensively improve the quality of power supply services,has become the consensus of various State Grid Corporations.This thesis takes a State Grid Corporation customer as the research object,"95598 historical work order" as the data basis,and applies data mining technology to study the complaints of power customers.The main research work is as follows:(1)In order to mine valuable features for the power customer complaint prediction model,a feature construction method is proposed.Analyze the influencing factors of power customer complaints,and combine the data characteristics and mining requirements of the "95598historical work order",put forward the idea of feature construction from four different perspectives of trajectory,time,business and customer sentiment,and then use this idea to design data mining technology Based on the feature construction method.(2)Aiming at the problem of serious imbalance between positive and negative samples in power customer complaint work orders,proposed to improve the random forest algorithm based on SMOTE and Bayesian optimization algorithm,namely the BSMOTE-RF algorithm.The classic data mining classification algorithm and the BSMOTE-RF algorithm are selected for experimental comparison and analysis.The results show that the classification accuracy of the BSMOTE-RF algorithm on large data sets and the handling of positive and negative sample imbalance problems are better than SVM,Naive Bayes and Random forest algorithm.(3)Use the BSMOTE-RF algorithm to design a power customer complaint prediction model and apply it to real power customer service data to predict whether power customers will initiate complaints in the future.Experiments have shown that the model can efficiently and accurately predict whether power customers will initiate complaints in the future,and can assist the State Grid Corporation in carrying out service work and improve customer satisfaction.The proposed feature construction method and BSMOTE-RF algorithm are more suitable for power customer complaint prediction,and the experiment proves that they are helpful to efficiently predict whether power customer will lodge a complaint.