User Location Prediction Based On Mobile Payment

In recent years,the smart devices develop rapidly.In the current conditions,it is a hot research direction to predict which room the users of smart devices are in.In this thesis,for the indoor environment under the coverage of multiple WIFI signals,along with the real environment of unstable factors such as position in motion and signal attenuation of WIFI equipment,we predict the accurate indoor room positioning of mobile phone users.This thesis takes shopping mall as the study case,and adopts big data of mobile payment using mobile phone to locate the user to the exact shop.With the popularity of smart phone,the O2O(Online to Offline)consumption model has developed rapidly.Accurate advertising in offline shop is an important marketing tool for O2 O,but for offline shop,random delivery will cause meaningless interference to most users in the mall,and also increase the cost of the merchant.Therefore,in the offline shopping malls,that using the information of mobile phone to predict which shop the mobile phone user is located in a real time manner is particularly important for the accurate delivery of advertisements.Due to the popularity of mobile payment,in the online shopping malls,users generate a large amount of mobile phone information and location tags in the Alibaba platform.Therefore,it is easy to collect the data and the amount of data is large.Moreover,the location information is accurate and does not need to be manually labeled.We use the method of big data and machine learning to mine it,and predict the shop which user is located in the mall,with real-time forecasting conditions and promotion conditions.Due to the large number of WIFI devices in the real environment of shopping mall,the signal transmission intensity and location are different,and the latitude and longitude information of the mobile phone might be inaccurate.Therefore,the mobile phone data has much noise,and it is difficult to locate the shop that the user might be in.Using more than 1 million real user behavior data for feature extraction and analysis,this thesis proposes a fusion model based on Bayesian algorithm for XGBoost and random forest.By training data of user's mobile phone,it predicts the location of users in the mall,which is the basis for decision making so as to provide accurate recommendation services for users in the future.This thesis deeply explores the information of user's mobile phone,constructs the feature of the user's geographic information,extracts the list of signal of WIFI detected by the user's mobile phone and the corresponding signal strength,the user's consumption location,the time stamp of consumption,and the shop's related characteristics such as business category.Using a variety of classification algorithms commonly used,such as logistic regression,SVM algorithm,XGBoost and random forest algorithm,we train the model with extracted user geographic information related features,and the accuracy of prediction by each model is compared.The performances of XGBoost algorithm and random forest are better than the other two algorithms for prediction which shop the user is in.In order to take the advantages of XGBoost algorithm and random forest algorithm,this thesis uses Bayesian algorithm to fuse the two models for predicting the shop where the user is in.Finally,we use the information of mobile phone and train the model.The accuracy of prediction is up to 90%,and each prediction only takes 0.63 ms.The result reaches the promotion accuracy and real-time prediction requirements.