Traffic Flow Prediction: An Intelligent Scheme for Forecasting Traffic Flow Using Air Pollution Data in Smart Cities with Bagging Ensemble

Noor Ullah Khan, Munam Ali Shah, Carsten Maple, Ejaz Ahmed, Muhammad Nabeel AsgharSustainability 2022 · DOI

TL;DR

This paper proposes a novel approach to traffic flow prediction in smart cities by integrating air pollution data with traditional traffic data. By employing a KNN bagging ensemble model, the researchers achieved a significant reduction in prediction error, demonstrating the value of air quality indicators for more accurate traffic forecasting.

What Problem Does This Paper Address?

Traffic congestion is a major issue in smart cities, leading to wasted time, fuel, and significant air pollution that harms public health. Accurate traffic flow prediction is crucial for effective traffic management, enabling drivers to choose optimal routes and reducing overall congestion. While many studies have explored traffic prediction using various machine learning techniques, few have effectively incorporated air pollution data as a predictive feature. This paper aims to bridge this gap by developing a more accurate traffic flow prediction system that leverages the correlation between air pollution and traffic congestion.

Methods

The study employs a multi-stage methodology. Initially, various simple regression techniques were evaluated to identify the best-performing individual models. Subsequently, ensemble techniques, specifically bagging and stacking, were applied to further enhance prediction accuracy. The core of the proposed method is a KNN bagging ensemble model. Bagging involves training multiple instances of the same base learner (in this case, KNN) on different bootstrap samples of the data and then aggregating their predictions. This approach is designed to reduce variance and improve the robustness of the model. The researchers also compared their ensemble models against individual regression models to demonstrate the benefits of their approach.

Datasets

The study utilized a large-scale, publicly available dataset from the CityPulse Aarhus, Denmark initiative. This dataset comprises two main components:

  • Pollution data: Including measurements of CO, SO2, O3, PM, NO2, latitude, and longitude.
  • Traffic intensity data: Providing information on traffic flow between two points.

Both datasets were collected in real-time and merged using timestamps. The paper notes that this dataset is widely used in the literature.

Key Insights and Ideas

  • Air pollution as a strong traffic predictor: The research highlights that air pollution data, particularly pollutants like CO, SO2, and NO2, are strongly correlated with traffic intensity. This suggests that air quality can serve as a valuable proxy for predicting traffic congestion.
  • Ensemble methods outperform individual models: The study demonstrates that ensemble techniques, especially bagging, significantly improve prediction accuracy compared to individual regression models.
  • KNN bagging excels: The K-Nearest Neighbors (KNN) algorithm, when combined with the bagging ensemble technique, emerged as the most accurate model for traffic flow prediction in this study. This indicates that KNN's ability to capture local patterns, when aggregated through bagging, is highly effective for this task.
  • Cost-effectiveness of air pollution sensors: By focusing on air pollution data, the approach offers a potential cost-saving alternative to deploying extensive and expensive traffic-specific sensor networks.

Key Results

The experimental results show that the KNN bagging ensemble model significantly outperformed other regression models and ensemble combinations. Specifically, the KNN bagging ensemble model reduced the error rate in predicting traffic congestion by more than 30% compared to previous studies using boosting. The paper reports that the KNN bagging model achieved the lowest RMSE (1.12) and MAE (0.55) values among the evaluated ensemble models.

Limitations and Future Work

  • Dataset Specificity: The study primarily used data from Aarhus, Denmark. The authors acknowledge that traffic patterns and pollution contributions can vary significantly across different geographical locations and seasons.
  • Seasonality and External Factors: Future work should investigate the impact of different seasons (e.g., summer vacations affecting traffic) and explore contributions to air pollution from other sources like home heating and equipment.
  • Electric Vehicle Impact: The increasing adoption of electric vehicles could influence future pollution patterns, posing a potential threat to the long-term validity of models that rely heavily on current emission trends.
  • Broader Data Inclusion: Incorporating data from more cities and exploring satellite measurements for traffic in conjunction with ground sensors are suggested for future research to enhance model robustness and generalizability.
  • Interpretability and Computational Cost: The paper notes that bagging can increase storage space and computation time, and may reduce model interpretability. These are areas for potential improvement.

Summary updated

Abstract

Traffic flow prediction is the most critical part of any traffic management system in a smart city. It can help a driver to pick the most optimized way to their target destination. Air pollution data are often connected with traffic congestion and there exists plenty of research on the connection between air pollution and traffic congestion using different machine learning approaches. A scheme for efficiently predicting traffic flow using ensemble techniques such as bagging and air pollution has not yet been introduced. Therefore, there is a need for a more accurate traffic flow prediction system for the smart cities. The aim of this research is to forecast traffic flow using pollution data. The contribution is twofold: Firstly, a comparison has been made using different simple regression techniques to find out the best-performing model. Secondly, bagging and stacking ensemble techniques have been used to find out the most accurate model of the two comparisons. The results show that the K-Nearest Neighbors (KNN) bagging ensemble provides far better results than all the other regression models used in this study. The experimental results show that the KNN bagging ensemble model reduces the error rate in predicting the traffic congestion by more than 30%.
BibTeX
@inproceedings{khan2022traffic,
  author = {Noor Ullah Khan and Munam Ali Shah and Carsten Maple and Ejaz Ahmed and Muhammad Nabeel Asghar},
  title = {{Traffic Flow Prediction: An Intelligent Scheme for Forecasting Traffic Flow Using Air Pollution Data in Smart Cities with Bagging Ensemble}},
  booktitle = {Sustainability},
  year = {2022},
  doi = {10.3390/su14074164},
}
Figure