A Framework of Travel Mode Identification Fusing Deep Learning and Map-Matching Algorithm

Abstract

The ubiquity of mobile phone signaling data (MPSD) allows us to study travel mode identification (TMI) of a larger scale of population in cities than GPS data and travel survey. Existing studies suffers from deficient data cleaning, ignoring low spatial accuracy of data and lack of the common datasets labeled with travel modes. This study proposes a novel TMI framework with MPSD and pseudo MPSD by designing hidden markov model (HMM)-based map-matching algorithm and bi-directional long short-term memory (Bi-LSTM)-based deep neural network (DNN). In this framework, conventional data cleaning method is modified to eliminate outliers in raw MPSD to improve data quality. In addition, HMM-based map-matching algorithm is designed to estimate the actual position of travelers from MPSD for spatial-temporal accuracy improvement. Specifically, pseudo MPSD is reconstructed by multi-source data aiming at overcoming the facts lacking of labeled MPSD. Both MPSD and labeled pseudo MPSD are processed to features, where the latter is used to train model based on Bi-LSTM, and the former is input to the trained model for TMI application validation. To evaluate the framework performance, a case study is carried out. Experimental results show that our framework outperforms baseline models on the same datasets. Furthermore, the accuracy improvement of designed approaches is demonstrated by a series of algorithm comparisons. The framework has good generality and flexibility, and can be extended to other cities for TMI.