Identifying Transportation Modes from Trajectory Dataset using Boosting and Deep Learning Methods in Intelligent Transportation Systems

  with the expansion of urbanization, the need for intelligent transportation to facilitate the movement of citizens has received more attention than before. Identifying and predicting the use of transportation modes is one of the most basic prerequisites for setting up and using intelligent transportation services. With the advancement of location technologies, agents, and smartphones, a lot of information is generated by many devices using Global Navigation Satellite Systems (GNSS). In this study, 4 Point features and 59 travel features and advanced features were extracted, four classification models GB, XGBoost, LightGBM and CatBoost subset of Boosting method after hybrid feature selecting with 3 classification models CNN, LSTM, and ConvLSTM subset of Deep Learning method Implemented and analyzed to predict transportation modes including walking, using bicycles, using buses, using cars, and using trains usage the GeoLife dataset. Finally, the LightGBM model predicted transportation modes with higher accuracy (95.49%) and less time complexity than other models.