ZCWarren/Global-Flight-Network-Analysis

NUS PC5253 Final Project

Global Flight Network Modeling & Analysis

The goal is to study the global flight network properties and investigate if they are related to or predictive of passenger numbers at different airports.

1. Research Questions and Objectives

• Main research questions:
  • What are the key properties of the global flight network?
  • How do these network properties relate to passenger numbers at different airports?
• Specific network properties to investigate(etc.):
  • Degree distribution
  • Average path length
  • Clustering coefficient
  • Centrality measures
• Objectives:
  • Identify the most significant network properties that predict passenger numbers
  • Develop predictive models to estimate passenger numbers based on network properties

2. Data Collection

• Data sources:
  • a. OpenFlights
  • b. SNAP: Stanford Network Analysis Project - Reachability
  • c. Air Transportation Networks
  • d. List of busiest airports by passenger traffic (2010-2015)
  • e. Top 20 busiest airports in the world
  • f. List of busiest airports by passenger traffic on Kaggle

3. Network Analysis

We employ the Kamada-Kawai layout algorithm to spatially visualize the network.

• Calculate basic network metrics:
  • Degree distribution
  • Average path length
  • Clustering coefficient
• Identify key airports (nodes) based on centrality measures:
  • Degree centrality
  • Betweenness centrality
  • PageRank
• Analyze network connectivity and resilience:
  • Giant component size
  • Percolation threshold

4. Machine Learning Analysis: Relations with Passenger Numbers

• Develop ML models:
  • Supervised linear regression
• Split data into training and testing sets
• Evaluate model performance:
  • Mean squared error
  • R-squared
• Perform Leave-One-Out Cross-Validation

5. Results Interpretation and Discussion

• Comprehensive Network Analysis: Analyzed the 50 busiest airports as of 2018, elucidating their structural and operational roles within the global flight network.

• ML Model Accuracy: The supervised linear regression model achieved up to 85.1% accuracy for predicting Degree and Betweenness centralities in both the validation set and unseen 2022 data.

• Key Predictors: PageRank emerged as a significant predictor of airport centralities, more so than passenger numbers, underscoring its strategic relevance.

• Impact of Passenger Numbers: Minimal impact of passenger numbers on predictions, affirming the stability of the global flight network and enhancing model generalization amid the dynamic conditions of 2022 driven by the COVID-19 pandemic.

• Strategic Insights: Highlighted the potential of combining network analysis and machine learning to develop robust, adaptable strategies for the aviation industry in an increasingly interconnected world.

Usage Statement

This project contains two branches:

1. main: This branch includes Python files that calculate basic network parameters.
2. Passenger-relations: This branch contains:
   • Passenger_network.ipynb: Investigates flight network parameter rankings for the top 50 airports by passenger numbers.
   • Another Jupyter Notebook file that includes the use of machine learning models to predict flight network parameters.

The interpretation rights of the research findings belong to the author. Other usage conditions adhere to the MIT License. If you use the code or conclusions from this project, please cite the source.

License

This project is licensed under the MIT License. See the LICENSE file for details.