Research And Implementation Of Location Prediction Method Based On Mobility Pattern And High-order Markov Model

In recent years,mobile devices have become popular,positioning technology has developed rapidly,and location information has become easier to acquire.Location-based services have become a common service in life.It can be divided into services based on current location and services based on future location.Among them,services based on future location need to more accurately predict the future location of users.It has important application value in the fields of information recommendation,spatial planning,and security supervision.Preprocessing of data is a necessary step in location prediction studies.Data is useful temporarily.Research with data that exceeds the valid time period will produce errors,so it is necessary to make data currency decisions in the data preprocessing.In addition,location data has complex temporal and spatial characteristics,so we needs to be modeled and analyzed.Markov model is effective in representing time series data,and it is simple and efficient,so it is widely used for location prediction.However,Markov model does not consider the influence of the historical state for the future state,so the accuracy of the result in the location prediction is not enough.It also has the problem that predictions cannot be made when the current state is not found in the historical data.This is also known as the zero frequency problem.In view of the above problems,this paper proposes to use the new stay point ratio to determine the currency of the data,and use the high-order Markov model for location prediction.In order to make full use of the user's historical data,the high-order Markov model which can continuously perform the reduced order operation is studied,and the temporal characteristics are comprehensively considered to improve the prediction accuracy.For the case where the current state is still not searched after the order is reduced multiple times,the location prediction is assisted by analyzing the user's mobility pattern.Finally,data currency determination and location prediction experiments were carried out on the actual dataset Geolife.The experimental results show that it is feasible and effective to use the new stay point ratio to determine the data currency.At the same time,the higher prediction accuracy can be obtained by using the variable order Markov model combined with time characteristics.Combined with the user mobility pattern analysis,the possible zero frequency problem can be better solved.