| In recent years,with the rapid spread of wireless networks and the widespread use of mobile intelligent terminals,the demand for location-based services(LBS)has grown rapidly.At present,LBS has rapidly developed and spread to all aspects of social life,and positioning technology has been closely linked with the development of LBS.With the continuous development of modern society,the urbanization process is accelerating,and the number of large buildings is increasing.More than 80% of the time is spent in indoor environments.The rapid spread of various mobile communication devices,catering,shopping,entertainment,and subway transportation have become important lifestyles for people,making indoor positioning and navigation an indispensable part of our lives.Due to the rapid development of Wi Fi technology,indoor positioning technology based on wireless local area network and signal reception strength utilizes the existing public WLAN infrastructure,does not require any other professional equipment,and only requires specific positioning software to achieve positioning.The WLAN-based indoor positioning technology has lower cost and can meet the requirements of indoor positioning for positioning accuracy,and has become a research hotspot.In this paper,based on the user's Wi Fi data and some location information of the e-commerce platform,the real scene of the data is restored to the maximum extent.After analyzing and processing the original feature group,the original data features are extracted and the key features affecting the positioning user are found.Information selection based AP selection algorithm.Extracting the cross-extracting feature groups that are rich and relevant to the business scene,in order to maximize the user's true behavior habits.Secondly,the feature selection based on the tree model reduces the data dimension and reduces the computational complexity.Get the sort result of the importance of the feature.Finally,the model is tested using different feature combinations,and the model is cross-validated using K-nearest neighbor algorithm,random forest and XGBoost.For KNN,you need to manually set the value of K.Using the pipeline method,you can input the k value at one time,which not only saves time,but also finds the best k value faster.The improvement to XGBoost is mainly to improve the regularization in the model of XGBoost and to adjust the parameters of the model to get the best experimental results.Experiments show that the combination of the proposed features and the improved XGBoost algorithm improves the accuracy and operational efficiency of the positioning user. |
