Inferring transportation modes from GPS trajectories using a convolutional neural network

Abstract

Identifying the distribution of users' transportation modes is an essential part of travel demand analysis and transportation planning. With the advent of ubiquitous GPS-enabled devices (e.g., a smartphone), a cost-effective approach for inferring commuters' mobility mode(s) is to leverage their GPS trajectories. A majority of studies have proposed mode inference models based on hand-crafted features and traditional machine learning algorithms. However, manual features engender some major drawbacks including vulnerability to traffic and environmental conditions as well as possessing human's bias in creating efficient features. One way to overcome these issues is by utilizing Convolutional Neural Network (CNN) schemes that are capable of automatically driving high-level features from the raw input. Accordingly, in this paper, we take advantage of CNN architectures so as to predict travel modes based on only raw GPS trajectories, where the modes are labeled as walk, bike, bus, driving, and train. Our key contribution is designing the layout of the CNN's input layer in such a way that not only is adaptable with the CNN schemes but represents fundamental motion characteristics of a moving object including speed, acceleration, jerk, and bearing rate. Furthermore, we ameliorate the quality of GPS logs through several data preprocessing steps. Using the clean input layer, a variety of CNN configurations are evaluated to achieve the best CNN architecture. The highest accuracy of 84.8% has been achieved through the ensemble of the best CNN configuration. In this research, we contrast our methodology with traditional machine learning algorithms as well as the seminal and most related studies to demonstrate the superiority of our framework. GPS is a ubiquitous positioning tool that records spatiotemporal information of moving objects carrying a GPS-enabled device (K. Heaslip). Transportation Research Part C 86 (2018) 360-371 0968-090X/ Published by Elsevier Ltd.T (e.g., a smartphone). The main advantageous of smart phones, compared to other GPS-equipped devices, is its enormous market penetration rate in a large number of countries and being relatively close to users nearly all of the time. As a consequence, such a dominant and area-wide sensing technology is capable of creating massive trajectory data of vehicles and people. A GPS trajectory, also called movement, of an object is constructed by connecting GPS points of their GPS-enabled device. A GPS point, here, is denoted as (lat, long, t), where lat, long, and t are latitude, longitude, and timestamp, respectively. The study of individuals' mobility patterns from GPS datasets has led to a variety of behavioral applications including learning significant locations, anomaly detection, locationbased activity recognition, and identification of transport modes (Lin and Hsu, 2014), in which the latter is the focus of this study. Nonetheless, GPS devices can only record time and positional characteristics of travels without any explicit information on utilized transport modes....

Ensemble Convolutional Neural Networks for Mode Inference in Smartphone Travel Survey

“…They took the average of the softmax class probabilities, predicted by each CNN model to generate the transportation label posteriors. Although the study carried out by Dabiri and Heaslip [10] used the CNN models, their study is different from our study in the following ways.…”

“…Furthermore, we have used a different ensemble method, i.e. a random forest model as a meta-learner explained in Section III, which demonstrates better prediction performance over the ensemble method developed by Dabiri and Heaslip [10].…”

“…Apart from the features used in the mode inference literature, features used in other fields of study may contribute to transportation mode detection [10]. "Jerk" has been used in traffic safety analysis to identify critical traffic events.…”

“…"Jerk" has been used in traffic safety analysis to identify critical traffic events. Jerk is defined as the derivative of acceleration, or the rate of change of acceleration over time [10]. GPS bearing rate has been used in driver behaviour profiling using smartphone data [11].…”

“…We considered five channels for the input data for the CNN models: distance to previous point, speed, acceleration, jerk, and bearing rate, as recommended in the literature [10]. Speed is calculated using the distance between each two consecutive GPS points divided by their time interval.…”