Research On Customer Group Prediction Based On Offline Retail

In recent years,the impact of e-commerce has made the traditional retail enterprises a great challenge.After being hit by the electricity supplier,physical retailers began to explore new marketing mode.In the field of data analysis,customer group prediction can provide many valuable information for retail businesses,so this paper studies the problem of customer prediction for offline retail.Based on the consumption data of ten thousand customers in a large shopping mall for three years and other brand category data,this paper classifies and predicts customer's sex and store by analyzing customer's behavior difference.In terms of gender classification,most of them are classified based on single attributes,such as sales data or names,and lack of comprehensive and comprehensive multi-attribute combination forecasting mode.Therefore,a fusion neural network model based on historical sales data,weather and name is proposed in this paper.The fusion network is a two independent model as the basic model:on the one hand,sales characteristics and weather features based on feature selection using chi square test and analysis method for dimensionality reduction using principal components,are classified by neural network;on the other hand,based on the name of features are classified by Bayesian method.Finally,the above two models are fused through neural networks.Experiments show that our proposed method is better than the traditional classification method,and the model fusion is better than the data fusion way,which satisfies the requirement of sex classification in offline retail application scenario.On the problem of customer back shop forecasting,there are the following problems in traditional processing methods:the selection of feature dimensions is mostly adopted by RFM model,although it has been recognized in academic and application aspects,but the model ignores the factors of customer's continuous purchasing ability;in data coding,the single encoding method is not perfect for the expression of data.In terms of model selection,the traditional machine learning algorithm can not extract high-dimensional features,while deep learning neural network needs a lot of data to improve performance.Based on the above problems,the work of this paper is as follows:on the basis of the RFM model of traditional customer classification,the first purchase time dimension is added,and the REFM feature model which is more suitable for the store problem is put forward.In the aspect of data coding,the feature subdivision method combining the number sharing and the K-means clustering algorithm is put forward,and the fusion is proved.The combination method is superior to the single feature processing method.In the model processing,a new fusion neural network prediction model is proposed by combining the AdaBoost and LSTM model methods.The experimental results show that our prediction model is superior to the traditional prediction model,and the research on the customer back shop problem satisfies the offline retail.It has certain application value for customer's forecasting demand.