Study of the US Road Network based on Social Network Analysis

Elie Ngomseu Mambou, Samuel Nlend, Harold Liu

Abstract

The complexity of big data structures and networks demands more research in terms of analysing and representing data for a better comprehension and usage. In this regard, there are several types of model to represent a structure. The aim of this article is to use a social network topology to analyse road network for the following States in the United States, US: California, Pennsylvania and Texas. Our approach mainly focuses on clustering of road network data in order to create "communities".

Introduction

This research explores the application of Social Network Analysis (SNA) techniques—tools typically reserved for mapping digital friendships and viral tweets—to the physical infrastructure of the United States. By treating road intersections as "nodes" and the roads themselves as "edges," we demonstrate how big data frameworks like Apache Spark can uncover hidden community structures and influential hubs within the transportation networks of California, Pennsylvania, and Texas.

Problem Statement

Analyzing massive road networks is computationally expensive and traditionally falls under the domain of Geographic Information Systems (GIS). However, GIS often focuses on spatial coordinates rather than the underlying topology—the way things are interconnected. As datasets grow into the "Big Data" realm, more robust ways are needed to understand network flow, identify bottlenecks, and group regions logically. By borrowing from SNA, this study aims to identify which intersections are the most "influential" for traffic flow and define "communities" based purely on road connectivity.

Methods and Tech Stack

The study utilizes a modern "Big Data" tech stack to process millions of data points. The road network is modeled as an undirected graph $P = (N, E)$, where $N$ represents intersections and $E$ represents the roads connecting them.

The Apache Spark ecosystem used for processing large-scale graph data.

The analysis was built on Ubuntu 14.04, using Spark 1.5.2. We leveraged GraphX for graph processing, Spark Streaming for real-time data handling, and MLlib for machine learning tasks like K-means clustering. Centrality metrics such as Degree, Closeness, and Betweenness were evaluated alongside the PageRank algorithm to measure node importance.

Datasets

The study used three primary datasets from the Stanford Large Network Dataset Collection (SNAP):

  1. roadNet-CA: California road network (1,965,206 nodes; 2,766,607 edges).
  2. roadNet-PA: Pennsylvania road network (1,088,092 nodes; 1,541,898 edges).
  3. roadNet-TX: Texas road network (1,379,917 nodes; 1,921,660 edges).

Key Insights

  • The "Social" Road: Roads behave like social networks. Certain intersections act as "hubs" with high betweenness that, if removed, would significantly disrupt the network.
  • Topological Similarity: Adjacency matrix visualizations across all three states showed a strong concentration along the diagonal. This suggests that despite different geographies, large-scale road networks follow a similar structural "signature" where urban cores are dense and peripheries are sparse.
  • PageRank vs. Degree: A node’s PageRank increases proportionally with its degree. Pennsylvania’s top intersections showed higher PageRank values than those in Texas, suggesting a more tightly integrated core network in PA.

Results

The researchers proved that the Spark framework is highly effective for iterative graph algorithms on millions of nodes. The paper successfully identified the top 10 most "important" nodes in the Pennsylvania network using PageRank.

Console output showing the top-ranked nodes in the Pennsylvania network based on PageRank scores.

As shown in the results, the highest-ranked node (ID: 859326) achieved a score of approximately 2.99. Additionally, K-means clustering (with $k=3$ and $k=4$) successfully partitioned the California dataset into distinct communities, with centroids representing the "heart" of each identified road region.

Conclusion and Future Work

While K-means provided a solid baseline for community detection, its rigidity regarding the number of clusters and Euclidean distance suggests that future work should explore more flexible methods like Fuzzy C-means or K-medoids. This SNA-based approach can be extended to traffic estimation and smarter urban planning, providing a topological perspective that complements traditional GIS analysis.

Summary updated

Abstract

The complexity of big data structures and networks demands more research in terms of analysing and representing data for a better comprehension and usage. In this regard, there are several types of model to represent a structure. The aim of this article is to use a social network topology to analyse road network for the following States in the United States, US: California, Pennsylvania and Texas. Our approach mainly focuses on clustering of road network data in order to create "communities".
BibTeX
@inproceedings{mambou2017study,
  author = {Elie Ngomseu Mambou and Samuel Nlend and Harold Liu},
  title = {{Study of the US Road Network based on Social Network Analysis}},
  booktitle = {AFRICON},
  year = {2017},
  url = {https://openreview.net/forum?id=wX2JzPaPNYD},
}
Figure