Forecasting the evolution of fast-changing transportation networks using machine learning.

Area of Science:

• Complex systems science
• Network science
• Machine learning applications

Background:

• Transportation networks are crucial for mobility and trade but contribute to disease spread and CO2 emissions.
• Understanding the dynamics of network changes is essential for infrastructure planning and policy development.
• Existing research often overlooks the predictive capabilities of machine learning in dynamic transportation systems.

Purpose of the Study:

• To investigate edge removal dynamics in mature, evolving transportation networks.
• To apply machine learning models for predicting monthly changes in these networks.
• To forecast the potential impact of CO2 emission reduction policies on the U.S. air transportation network.

Main Methods:

• Utilized machine learning algorithms to predict edge removals in the Brazilian domestic bus and U.S. domestic air transportation networks.
• Trained and tested models on monthly data to assess predictive accuracy within and across different timeframes.
• Developed a forecasting approach to simulate the effects of significant reductions in the U.S. air transportation network scale.

Main Results:

• Machine learning models achieved high accuracy in predicting same-month edge removals for both networks.
• Models trained for a specific month demonstrated accuracy for other months in the U.S. air network, even with external shocks.
• The forecasting approach successfully simulated the impact of hypothetical CO2 reduction policies on air travel.

Conclusions:

• Machine learning offers a powerful tool for understanding and predicting the evolution of complex transportation networks.
• The predictive accuracy suggests potential for proactive infrastructure planning and policy impact assessment.
• This methodology can aid in developing future scenarios for sustainable transportation infrastructure development.