An Application And Analysis Of Forecast Custom Churn Based On KELM-AE Algorithms Model

At present,in the context of business diversification in various industries,market saturation,and the trend of anti-globalization due to the epidemic and changes in the international political environment,competition between domestic and foreign companies has become increasingly fierce.Moreover,the global economic environment is still in the aftermath of the impact of the epidemic.These comprehensive factors have led to increasingly high costs for companies to attract new customers.All companies are shifting from incremental operations to stock operations.How to better retain old customers has become a concern for companies.The key point is that the problem of customer churn prediction has therefore become a hot issue that the industry pays close attention to.However,customer data information has the characteristics of high complexity,high redundancy,and high dimensionality.How to build a reliable and accurate prediction model and predict possible customer churn behavior through technical means has become an important research in industry and academia.Subject.This paper studies the problem of customer churn prediction,and proposes a customer churn prediction algorithm based on an improved autoencoder and a customer churn prediction algorithm fused with convolutional neural networks.The research results are as follows:(1)A customer churn prediction algorithm based on an improved autoencoder is proposed.This model addresses the increase in calculations caused by the design of too many hidden layer nodes and the complexity of customer feature data when the autoencoder is processing large-scale discrete data feature extraction,and an extreme learning machine is added to the autoencoder.,The use of extreme learning machine only needs to set the number of initial hidden layer nodes,without dynamically adjusting the weight and bias of the network input,the characteristics of the global optimal solution can be obtained,and the kernel function is added to optimize the extreme learning machine susceptible to extreme randomness The problem of the influence of the set value not only retains the advantages of the unsupervised learning of the self-encoding network itself,but also improves the ability to extract discrete data features and the robustness of feature expression,while greatly reducing the amount of calculation.(2)On the basis of the kernel extreme learning machine-autoencoder model,aiming at the problem of poor training effect of shallow BP neural network,a kernel extreme learning machine-autoencoder(KELM-AE-CNN)integrated into the convolutional neural network is proposed Customer churn prediction algorithm.Utilize the advantages of the kernel extreme learning machine-autoencoder model in processing discrete feature data,replace the BP neural network with a convolutional neural network,use the advantage of the convolutional neural network to process high-dimensional feature vectors,and use continuous feature data After the normalization is processed and the discrete feature data after feature extraction are combined as the input of the convolutional neural network,the accuracy of the neural network model is increased on the basis of ensuring the accuracy of feature extraction.In this paper,the two algorithm models are compared with a variety of algorithm models,and have achieved good prediction performance on two public customer churn data sets,which has certain advantages.