Early Transportation Mode Detection Using Smartphone Sensing Data

Abstract

The advancement in sensing technology has enabled the development of various applications for activity recognition using smartphone sensor data. One of the useful applications in an intelligent transportation system is the identification of transportation mode to provide context-aware assistance for the execution of systems such as driver assistant. Such real-time critical systems demand the early detection of transportation mode for making effective decisions. This paper proposes a method to detect the transportation mode at an early stage by achieving a decent trade-off between accuracy and earliness based on partially observed sensory time series data. As a result, a hybrid deep learning classifier is developed by utilizing the capabilities of the convolutional neural network, recurrent neural network, and deep neural network to learn the hidden temporal correlation of pattern information for the sensory data. In addition, a decision policy is defined on top of the classifier to perform the transportation mode prediction for the incoming time series by attaining acceptable trade-off. The proposed model is evaluated using two publicly available supervised datasets and demonstrated good performance in terms of accuracy and earliness. Also, the model is compared with the existing alternative for verifying the effectiveness.