Inferring transportation modes from GPS trajectories using a convolutional neural network (1804.02386v1)

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.

• The paper proposes a novel CNN-based model that directly uses raw GPS data (speed, acceleration, jerk, bearing rate in a 3D format) to classify travel modes, avoiding traditional feature engineering.
• The research achieved a high classification accuracy of 84.8% with their CNN model, significantly outperforming traditional machine learning methods and previous studies on the GeoLife dataset.
• Findings suggest CNNs offer substantial improvements in transport mode detection accuracy, enabling more automated urban planning and traffic management without intensive manual feature engineering.

An Overview of Inferring Transportation Modes from GPS Trajectories Using CNNs

The research article by Dabiri and Heaslip investigates the application of Convolutional Neural Networks (CNNs) for inferring transportation modes from raw GPS trajectories. Traditional methods for travel mode identification typically involve extensive feature engineering and are often subject to traffic and environmental conditions, as well as human bias. This paper presents a novel CNN-based model that directly leverages raw GPS data to predict travel modes including walk, bike, bus, driving, and train, showcasing an advanced methodological step in transportation mode inference.

Methodology and CNN Architecture

The core contribution of this paper lies in the novel design of the CNN's input layer and the use of deep learning to bypass traditional feature extraction processes. By structuring raw GPS data into a 3D format suitable for CNNs, they combined four channels of kinematic features: speed, acceleration, jerk, and bearing rate. These inputs are then fed into a variety of CNN architectures to develop an effective model for travel mode classification.

Data preprocessing involved cleansing and structuring GPS logs, detecting outliers, and applying smoothing techniques to refine the signal quality. The CNN architectures utilized include several convolutional layers, pooling layers to downsample features, fully connected layers for learning complex interactions, and dropout layers to address overfitting.

The research achieved a highest classification accuracy of 84.8% through an ensemble approach, emphasizing the effectiveness of their model over traditional algorithms and seminal studies in the transport mode inference field.

Comparative Analysis and Results

Their CNN framework was compared with conventional machine learning models such as K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Decision Trees (DT), Random Forests (RF), and Multilayer Perceptron (MLP). The CNN demonstrated a significantly higher test accuracy, averaging 84.8% compared to the best performance among traditional methods (78.1% by RF). The CNN advantage was further evidenced in outperforming the classical approaches, underscoring the powerful feature representation and learning capabilities of deep learning.

In a broader context, the CNN model proposed by Dabiri and Heaslip surpassed previous studies utilizing the Microsoft GeoLife dataset, showing an improvement exceeding 8% when compared with the traditional feature-based inference models and even outperforming recent deep learning attempts by up to 16%.

Implications and Future Research Directions

The findings indicate that CNNs can offer substantial improvements in transportation mode detection accuracy, paving the way for more automated and unbiased transport analytics without intensive manual feature engineering. This advancement has significant implications for fields dealing with urban planning, traffic management, and automated transport systems, potentially leading to smarter, data-driven infrastructural developments and policy-making.

Future research might explore enlarging the dataset, possibly employing semi-supervised and unsupervised methods to harness the abundant yet unlabeled GPS data. Additionally, integrating more diverse environmental data and expanding modes of transport could enhance model robustness and applicability across different urban contexts.

In conclusion, the research by Dabiri and Heaslip illustrates a notable advancement in travel mode inference, underlining the transformative potential of deep learning architectures like CNNs in deciphering complex behavioral data embedded within GPS trajectories.