Design And Implementation Of Telecom Customer Churn Prediction Model Based On Particle Swarm Optimization Algorithm

Under the overall environment of "number portability" and "faster and more affordable internet connection" proposed by the Central Government to be fully implemented among three major telecommunications carriers,the competition among telecommunications carriers must be white-hot.Therefore,incremental operation transforming to stock operation is the strategic transformation measure to be considered by the three major earriers.The value of stock users is far higher than that of new users.As a result,it is especially important to increase the value of stock users and reduce customer churn.The most critical measure is to establish a customer churn early-warning model,to predict the potential customer churn among stock users and take personalized marketing measures to pay a return visit on customers,which can effectively reduce the stock customer churn.Under this background,in this thesis the customer data on the big market of telecommunications carriers of a province has been taken as the study object,and three prediction models have been established through Python for comparison based on the customer churn prediction model at home and aboard.The main study completed by the author is as follows:(1)For data pre-processing,the author has mainly taken the true and desensitized data sets on the big market of telecommunications carriers of a province as study object,used data cleaning,the data equalization based on AdaCost algorithm,and feature selection based on ReliefF(Relief filtering)method,to ensure high-quality experimental data sets are obtained.(2)For modelling and experimental verification,the author has mainly introduced how to establish the customer churn prediction model with Bayesian algorithm,BP neural network algorithm,and particle swarm optimization,to predict telecommunications customers in class,and realized the three prediction models through the programming language Python.The experimental results show that the prediction model based on PSO-BP network is of higher prediction precision and faster network convergence rate compared with BP neural network prediction model,and of higher accuracy rate compared with Bayesian model.The result also shows that the superiority of neural network prediction model based on particle swarm optimization algorithm.The innovation of this thesis is to optimize the traditional neural network by particle swarm optimization algorithm,and process data sets by the feature selection based on Relief algorithm,to reduce irrelevant feature attributes and greatly improve the working efficiency of neural network.The study results show that the telecommunications customer churn prediction model based on particle swarm optimization algorithm is over 97%on the accuracy rate,and Garson algorithm is used for sensitivity analysis to output the impact factor of each feature attribute.Such model is commercially feasible for the management of carriers' customer churn and can be instructional for the practical management of stock customers of telecommunications carriers in the future.