Research Of Complaint Prediction In Mobile Networks With Data Mining

In recent years,with the rapid development of the mobile networks and intelligent terminal,many kinds of mobile internet application appear and have good development.With the diversity of business and the increasing number of users,the frequency of complaints in the telecommunications industry is also growing rapidly.User complaint handing is especially important for mobile operators in mobile networks and is the important way to improve the quality of mobile networks.In this situation,maintenance and operation of network,reduction of user complaint,has become the major works of mobile operators.At the same time,with the rapid development of information technology,the mobile operators have a large amount of data.The explosion of data contains a number of important information,but it is lack of effective use of these data information.User complaint is an important information source for network optimization work in mobile network operators.How to find the problem from the complex complaint data is a new problem faced by mobile operators.Therefore,we plan to use data mining technology and take advantage of data to build prediction model which predicts bursting complaints before it happens.Users' complaints in general reflect their dissatisfaction with the mobile network.Uncertain user behavior has greatly impact on the correlation between user complaint and the system abnormality.In this paper,we propose a prediction system based on network operational data and historical complaint data.With the help of data mining,the relationship between network anomaly and user complaints can be modeled,and predict whether a specific type of complaint burst will happen during a future time window.The operation of the prediction system consists of two parts: data preprocessing and prediction model building.First,two different kinds of network operational data,the performance data and anomaly alarm data,are collected and linked with complaint data for further preprocessing to generate original training set.Second,the random forest algorithm is applied on the training data set for model generation.The system is evaluated with real data collected from a major Chinese mobile operator.And we experiment with the parameters which have the most significant influence on system performance.These parameters include feature time window,prediction time window,resample ratio and random forest algorithm.The evaluation results show that the precision of the complaint bursts prediction can be as high as 80%,and the recall of the complaint bursts prediction can also achieve 60%.