Location Selection And Sales Forecast Research On Unmanned Retail Store Based On Machine Learning

Store location selection and initial operation after site selection are important decisions in store management.With the development of China's retail industry,there have been features of unmanned,new stores expanding at a fast rate and a high proportion.Unmanned retail liberates labor cost constraints in site selection research,making more flexible constraint combinations and more detailed feature granularity investigation possible.At the same time,the rapid expansion of new stores puts higher demands on the data collection speed and model decision speed in site selection decisions.However,at present,most of the site selection studies use manual collection,questionnaires and other non-automated methods for data collection and analysis;Delphi method,AHP and other traditional methods are used for site selection decisions.The data collection and model decision-making of these studies are slow,and they can no longer meet the requirements of rapid location selection in the new format.Therefore,in response to the requirements of rapid data collection and rapid analysis,this paper proposes an automated data processing framework based on the Google map system.The framework can automatically collect and mine geographic information data from the Google map system and convert it into structured data that can be understood by the site selection model.Aiming at the requirement of site selection decision speed,this paper uses machine learning model to make site selection decision,which strikes a balance between the interpretability of the decision result,the decision speed and the accuracy of the decision result.It also uses management theory to systematically analyze the decision logic of the machine learning model,and establishes an interpretable link between management and machine learning.The rapid expansion of new stores brings the characteristics of a high proportion of new stores.However,the current research on sales forecasting mainly focuses on the research of mature stores,and there are few researches on sales forecasting of new stores.Because there are few operating data for new stores,it is very difficult to predict sales in the absence of data.In response to this problem,this paper proposes a pattern recognition method based on sales characteristics.This method can find mature stores with the same sales characteristics as new stores from mature stores.Further,this paper designs a deep transfer learning model that can migrate the sales data of these mature stores to new stores and make predictions,so that the accuracy of sales prediction of new stores is close to that of mature stores.The research results of this paper provide an effective solution to the lack of operational data in the initial operation of retail stores,and provide a way of thinking for enterprise operation management.