Accurate Traffic Flow Prediction in Heterogeneous Vehicular Networks in an Intelligent Transport System Using a Supervised Non-Parametric Classifier

Hesham El-Sayed, Sharmi Sankar, Yousef-Awwad Daraghmi, Prayag Tiwari, Ekarat Rattagan, Manoranjan Mohanty, Deepak Puthal, Mukesh Prasad2018 · DOI

TL;DR

This paper tackles the challenge of accurately predicting traffic flow in dynamic heterogeneous vehicular networks (HETVNETs) to improve Quality of Service (QoS). The authors propose a refined Support Vector Machine (SVM) model using a Radial Basis Function (RBF) kernel, which demonstrates superior prediction accuracy and lower computational complexity compared to existing methods.

What Problem Does This Paper Address?

Intelligent Transportation Systems (ITSs) rely heavily on Heterogeneous Vehicular Networks (HETVNETs) for services like safety alerts. These networks, evolving from VANETs, face challenges due to the highly dynamic nature of vehicles, leading to intermittent connectivity and potential data loss or delays. Ensuring uninterrupted and timely service delivery requires accurate prediction of traffic conditions. While various prediction methods exist, there's a need for a classifier that can handle diverse traffic conditions, achieve high accuracy, and operate with low computational overhead for real-time applications.

Methods

The core of this work is a refined Support Vector Machine (SVM) model. SVMs are non-parametric supervised learning algorithms known for their ability to map non-linear data into high-dimensional feature spaces, creating non-linear decision boundaries. The authors specifically employ an RBF kernel, which is known for its high performance in both linear and non-linear SVMs.

The proposed model involves several stages:

  • Dataset Initialization: The HETVNET traffic data is pre-processed into an SVM-compatible format, converting categorical features to numerical ones.
  • Classification (Training and Testing): The paper explores two approaches: a traditional train-test split and a cross-validation procedure. Cross-validation is favored for its ability to reduce variance and prevent overfitting, especially with limited datasets.
  • SVM Model: The SVM with an RBF kernel is trained using the prepared data. The paper highlights the selection of the RBF kernel due to its balance of complexity, optimality, and accuracy, especially for datasets with a moderate number of features.
  • Attribute Selection: A correlation feature evaluator is used to rank and select the most relevant attributes for prediction, identifying "bytes received" and "signal strength" as highly influential.
  • Performance Measurement: Standard metrics like accuracy, precision, recall, F-measure, and Matthew's Correlation Coefficient (MCC) are used to evaluate the model's performance.

Datasets

The study utilizes a HETVNET traffic dataset from the Cambridge University repository (CRAWDAD). This dataset focuses on vehicle-to-road station unit (V2RSU) traffic, specifically segregated into northbound and southbound directions over five laps. The data includes various traffic characteristics such as packet sequence number, time, bytes sent/received, sender/receiver location and speed, and signal strength.

Key Insights and Ideas

  • Non-parametric methods excel in dynamic environments: The paper reinforces the idea that non-parametric methods, like SVMs, are more effective than parametric ones for traffic flow prediction due to their ability to capture the inherent non-linearity of traffic dynamics.
  • RBF kernel is a strong choice for HETVNETs: The RBF kernel for SVMs offers a good balance of performance metrics, making it suitable for the complex and dynamic nature of vehicular networks.
  • Cross-validation is crucial for robust prediction: Employing cross-validation significantly improves prediction accuracy and prevents overfitting, especially when dealing with datasets that might not be sufficiently large or representative.
  • Feature selection matters: Identifying and prioritizing key attributes like "bytes received" and "signal strength" can lead to more efficient and accurate models.

Key Results

The proposed RBF-refined SVM model achieved a remarkable accuracy of over 99% on the test dataset. This performance significantly outperformed traditional parametric classifiers like logistic regression (around 66% accuracy) and other SVM variants like LIBSVM and LIBLINEAR, especially when using the cross-validation approach. The paper also notes that the RBF kernel-based SVM demonstrated lower computational complexity compared to some other kernel types, making it more suitable for real-time applications.

Limitations and Future Work

The authors acknowledge that the dataset used was of moderate size. They suggest that in the future, a larger dataset could be used to further test the scalability and efficiency of the SVM-RBF model. Additionally, they propose developing an adaptive routing protocol for HETVNETs that leverages the proposed prediction model to mitigate issues like broadcasting storms and intermittent connectivity.

Summary updated

BibTeX
@misc{elsayed2018accurate,
  author = {Hesham El-Sayed and Sharmi Sankar and Yousef-Awwad Daraghmi and Prayag Tiwari and Ekarat Rattagan and Manoranjan Mohanty and Deepak Puthal and Mukesh Prasad},
  title = {{Accurate Traffic Flow Prediction in Heterogeneous Vehicular Networks in an Intelligent Transport System Using a Supervised Non-Parametric Classifier}},
  year = {2018},
  doi = {10.3390/s18061696},
}
Figure