The Research Oflocation Prediction Modeling Based On Mobile Phone Data

The scope of User’s activity places always shows his hobbies, activity pattern and personal social relationship. If the user’s activity prediction prototype can be established, it can locate the user’s next activity place in accordance with the present spot. The smart phone has already become a necessary part of people, thus establishing activity prediction prototype based on the research on the data of user’s terminal phone end is further meaningful.The article deeply discusses the human beings activity pattern model and Data Manning algorithms for prediction techniques and other relevant theory. And the thesis fully concludes the present research result of human beings activity pattern home and abroad. Since people’s activity pattern possessed the characteristics of chronological order for frequent visit spot, the article design and realize smartphone users’ activity prediction prototype based on the both of the revised Apriori calculating’s relevant spot prototype and revised PreFixSpan calculatingthe order of the locations prediction prototype.Firstly the history data is analyzed by the revised Apriori algorithm and the close correction of locations that the locations are always visited by the smartphone user on the same day are extracted from the history data sets. Secondly the motion trail of the user is generated by the revised PreFixSpan algorithm. During the prediction phase input the current location of the user the prediction model will match the location from the projected data base andfind alocation where the user will to visit next time.In order to create the prediction model, the thesis bases on the data from Nokia company to create prediction model. Use the first half year’s data as training data to create the model and the second half year’s data as test data. The accuracy of the model is about65%. The average accuracy of Nokia Big Data Challenge is59%. During the creating model phase, this paper uses a user’s data as training data to create prediction model, but the model have universality. The relative parameters and thresholds can be used for other users.