Prediction and Detection of Urban Trajectory Using Data Mining and Deep Neural Network

Abstract

With the development of modern cities, people’s traffic behaviors are on an ever-more increase. However, urban traffic is often congested due to bad road conditions or unreasonable road planning. To predict the trajectory of urban taxis, the taxi trajectories data are excavated based on the analysis of urban taxi behaviors. Then, the STTM (Spatio-Temporal Trajectory Model) is proposed using the LSTM (Long Short-Term Memory) and network residual. Meanwhile, the taxi is used for the urban road traffic perception and extraction, so that the taxi sensor and road traffic information can cooperate to construct a model for urban scale calculation. Afterward, the influence of the same model on the travel time predictions is compared for different urban roads. The results show that, based on the proposed STTM, the MRPE (Mean Relative Percentage Error) of the predicted value is 6.126%, the MAE (Mean Absolute Error) is 72.416 seconds, the MRE (Mean Relative Error) is 7.022%, the RMSE (Root Mean Square Error) is 293.977 seconds, and the coefficient of determination R2 is 0.884. This indicates that the model results have high goodness of fit, so it is a successful case of the application of urban computing in taxi trajectory prediction. Overall, the taxi ID (Identifier) and weather conditions have a great influence on the prediction results of urban taxi trajectory, and the STTM has a more obvious effect on improving the accuracy of travel time prediction for urban roads.