DeepTravel: Online transport mode identification based on low-rate sampling sensors through deep neural network

Transportation Mode Identification (TMI) has been studied for over 10 years, but few researches focused on datasets sampled at low sampling rates with low localization rates as a result of energy-efficient data collection. To fill this gap, we consider such a topic in this paper and present DeepTravel, an online TMI algorithm using a deep neural network with only ten features for each data sample, to classify four transportation modes: walking/stationary, bus, car and metro. We trained and evaluated our model on a dataset with an average sampling rate of 0.067Hz and localization rate less than 50%, and still managed to achieve an overall classification accuracy of 87%, which indicates its effectiveness in energy-efficient TMI tasks.