Preliminary Results on Classifying the RTK Trajectories Using the BP Neural Network

The classification of network RTK trajectories is significant to revealing the user behavior pattern and improving the services and managements. In this paper, a novel approach based on BP neural network is presented for classification on 5 types of behaviors as control point, line style, surface style, account sharing and invalid surveys. With the feature set consists 26 profiles, the BP neural network is trained and established. The preliminary results show that, the accuracy on training and testing data sets are respectively 84.3% and 81.2%. Especially for the account sharing and invalid activities, the accuracy reaches 94.7% and 93.8%, which indicates the well generalization ability. Moreover, the future work would be focus on improving the volume of data, optimizing trajectory shape feature indexes.