End-to-end trajectory transportation mode classification using Bi-LSTM recurrent neural network

Hongbin Liu, Ickjai Lee

Abstract: Transportation mode classification is a key task in trajectory data mining. It adds human behaviour semantics to raw trajectories for trip recommendation, traffic management and transport planning. Previous approaches require heavy pre-processing and feature extraction processes in order to build a classifier, which is complicated and time-consuming. Recurrent neural network has demonstrated its capacity in sequence modelling tasks ranging from machine translation, speech recognition to image captioning. In this paper, we pro­pose a trajectory transportation mode classification framework that is based on an end-to-end bidirectional LSTM classifier. The proposed classification process does not require any feature extraction process, but automatically learns features from trajectories, and use them for classification. We further improve this framework by feeding the time interval as an external feature by embedding. Our experiments on real GPS datasets demonstrate that our approach outperforms existing methods with regard to AUC.