[Paper Review] Transportation Modes Classification Using Feature Engineering

Mohammad Etemad|arXiv (Cornell University)|Jul 28, 2018

Human Mobility and Location-Based Analysis46 references4 citations

TL;DR

This paper proposes a feature engineering-based framework for transportation mode classification from GPS trajectories, using wrapper feature selection and cross-validation to optimize performance. It achieves superior accuracy compared to deep learning baselines, though ground truth noise removal requires test labels—limiting real-world applicability.

ABSTRACT

Predicting transportation modes from GPS (Global Positioning System) records is a hot topic in the trajectory mining domain. Each GPS record is called a trajectory point and a trajectory is a sequence of these points. Trajectory mining has applications including but not limited to transportation mode detection, tourism, traffic congestion, smart cities management, animal behaviour analysis, environmental preservation, and traffic dynamics are some of the trajectory mining applications. Transportation modes prediction as one of the tasks in human mobility and vehicle mobility applications plays an important role in resource allocation, traffic management systems, tourism planning and accident detection. In this work, the proposed framework in Etemad et al. is extended to consider other aspects in the task of transportation modes prediction. Wrapper search and information retrieval methods were investigated to find the best subset of trajectory features. Finding the best classifier and the best feature subset, the framework is compared against two related papers that applied deep learning methods. The results show that our framework achieved better performance. Moreover, the ground truth noise removal improved accuracy of transportation modes prediction task; however, the assumption of having access to test set labels in pre-processing task is invalid. Furthermore, the cross validation approaches were investigated and the performance results show that the random cross validation method provides optimistic results.

Motivation & Objective

• Address the challenge of accurately classifying transportation modes from GPS trajectory data using traditional machine learning.
• Improve classification performance by selecting optimal trajectory features through wrapper search and information retrieval methods.
• Investigate the impact of ground truth noise removal on model accuracy, acknowledging its dependency on labeled test data.
• Evaluate cross-validation strategies to avoid overly optimistic performance estimates in trajectory classification tasks.

Proposed method

• Apply wrapper feature selection to identify the most predictive subset of trajectory features from GPS records.
• Use information retrieval techniques to rank and select features based on their relevance to transportation mode prediction.
• Train and evaluate multiple classifiers on the selected feature subsets to determine the optimal combination.
• Implement cross-validation with random shuffling to assess model robustness, while identifying potential overfitting in results.
• Integrate ground truth noise removal during pre-processing, assuming access to labeled test data—a limitation for real-world deployment.
• Compare the final framework’s performance against two deep learning-based approaches from related literature.

Experimental results

Research questions

• RQ1 Can a carefully engineered feature set outperform deep learning models in transportation mode classification from GPS trajectories?
• RQ2 How does ground truth noise removal affect the accuracy of transportation mode prediction, and what are its practical limitations?
• RQ3 Which cross-validation strategy provides the most reliable performance estimation for trajectory-based classification tasks?
• RQ4 What is the optimal combination of features and classifiers for maximizing prediction accuracy in transportation mode detection?

Key findings

• The proposed feature engineering framework achieved higher classification accuracy than two deep learning-based approaches reported in related work.
• Ground truth noise removal improved prediction accuracy, but required access to test set labels—making it impractical for real-world deployment.
• Random cross-validation produced overly optimistic performance estimates, suggesting caution in model evaluation for trajectory data.
• Wrapper feature selection effectively identified a high-performing subset of trajectory features, enhancing model generalization.
• The study highlights the importance of data quality and evaluation methodology in trajectory-based machine learning tasks.
• The results suggest that traditional machine learning with strong feature engineering can outperform deep learning in this specific application domain.