Machine learning analysis of the effects of COVID-19 on migration patterns

Farzona Mukhamedova, Ivan TyukinScientific 2024 · DOI

Abstract

This study investigates the impact of the COVID-19 pandemic on European tourist mobility patterns from 2019 to 2021 by conceptualizing countries as monomers emitting radiation to model and analyze their patterns through the lens of socio-economics and machine learning. By incorporating perturbations into clustering, this work evaluates the stability of mobility flux clustering under variable conditions, offering insights into the dynamics of socio-economic corridors. The results highlight distinct shifts in tourist behavior, with bimodal clustering in 2019 reflecting heterogeneous mobility patterns, whereas unimodal distributions in 2020 and 2021 indicate increased global uniformity, driven by pandemic-induced restrictions and gradual recovery. The PCA and dendrograms of the perturbed clustering reveal that tourist preferences align with GDP, cultural, and linguistic similarities, explaining regional cohesion and fragility. This study demonstrates the fragility of emerging socio-economic corridors like the Red Octopus compared to the resilience of established ones like the Blue Banana. The findings emphasize the importance of targeted policy interventions, such as strengthening transport infrastructure and fostering small and medium-sized enterprises (SMEs), to mitigate disruptions and promote balanced regional development. By integrating perturbations into clustering, this research introduces a strong framework for assessing mobility patterns under realistic variability to enhance economic resilience and anticipate shifts in tourist behavior during global crises.

Introduction

The COVID-19 pandemic caused unprecedented disruptions to global human mobility. While epidemiological studies have extensively covered the virus's spread, the socio-economic impact on international travel patterns remains a critical area of inquiry. This study investigates how the pandemic restructured European tourist mobility from 2019 to 2021. By conceptualizing countries as "monomers" emitting radiation, the research models mobility flux to understand the stability of economic corridors under extreme stress.

Methodology

The researchers employed a multi-faceted approach combining physics-inspired modeling with machine learning:

  • Radiation Model: Borrowed from diffusion dynamics, this model predicts mobility flux between locations based on population density and the "radius" of surrounding populations. It treats countries as sources emitting radiation (tourists).
  • Perturbation-Enhanced Clustering: This is the paper's core methodological contribution. Instead of running a single clustering algorithm, the authors performed 100,000 Monte Carlo simulations, introducing controlled "noise" (perturbations) into their distance matrices to see which country clusters remained stable under variable conditions.
  • Multi-Metric Evaluation: They compared several clustering algorithms (K-means, DBSCAN, and Hierarchical Agglomerative Clustering) across different similarity measures like Cosine similarity, Pearson correlation, and Manhattan distance.
  • Data Sources: The study utilized empirical tourist flow data from the World Tourism Organization’s (UNWTO) Yearbook of Tourism Statistics for the years 2019–2021 and synthetic mobility flux generated by the Radiation Model.

Key Insights: From Heterogeneity to Uniformity

A primary finding of the study is the shift from "bimodal" to "unimodal" travel distributions. In 2019, European travel was characterized by a high degree of heterogeneity, with diverse motivations and destinations creating complex networks.

European tourist mobility flux in 2019, showing a heterogeneous, bimodal distribution of travel patterns.

With the onset of the pandemic in 2020, these patterns collapsed into a forced global uniformity. Travel restrictions effectively erased emerging socio-economic corridors, such as the "Red Octopus" in Southern and Eastern Europe, while established corridors like the "Blue Banana" (stretching from North West England to Northern Italy) demonstrated greater resilience.

European tourist mobility flux in 2020, illustrating the collapse of emerging corridors and a shift toward unimodal uniformity due to pandemic restrictions.

Principal Component Analysis (PCA) further revealed that tourist preferences are deeply rooted in GDP, cultural ties, and linguistic similarities. The breakdown of these traditional drivers in 2020—where the first principal component's variance explanation dropped from 92% to 91%—reflects the transition to "emergency logic" in mobility.

Conclusion and Policy Implications

The study highlights the fragility of newer economic networks compared to established ones. By 2021, travel patterns began to stabilize, achieving high silhouette scores (0.841 for empirical data), suggesting a recovery that follows population-based logic but remains more localized than pre-pandemic levels.

To enhance regional resilience, the authors recommend targeted policy interventions, including the strengthening of transport infrastructure and support for Small and Medium Enterprises (SMEs) within vulnerable corridors. Future work aims to evolve the "monomer" model into a "polymer" framework, allowing for higher spatial resolution and better modeling of border-zone radiation.

Summary updated

Abstract

This study investigates the impact of the COVID-19 pandemic on European tourist mobility patterns from 2019 to 2021 by conceptualizing countries as monomers emitting radiation to model and analyze their patterns through the lens of socio-economics and machine learning. By incorporating perturbations into clustering, this work evaluates the stability of mobility flux clustering under variable conditions, offering insights into the dynamics of socio-economic corridors. The results highlight distinct shifts in tourist behavior, with bimodal clustering in 2019 reflecting heterogeneous mobility patterns, whereas unimodal distributions in 2020 and 2021 indicate increased global uniformity, driven by pandemic-induced restrictions and gradual recovery. The PCA and dendrograms of the perturbed clustering reveal that tourist preferences align with GDP, cultural, and linguistic similarities, explaining regional cohesion and fragility. This study demonstrates the fragility of emerging socio-economic corridors like the Red Octopus compared to the resilience of established ones like the Blue Banana. The findings emphasize the importance of targeted policy interventions, such as strengthening transport infrastructure and fostering small and medium-sized enterprises (SMEs), to mitigate disruptions and promote balanced regional development. By integrating perturbations into clustering, this research introduces a strong framework for assessing mobility patterns under realistic variability to enhance economic resilience and anticipate shifts in tourist behavior during global crises.
BibTeX
@inproceedings{mukhamedova2024machine,
  author = {Farzona Mukhamedova and Ivan Tyukin},
  title = {{Machine learning analysis of the effects of COVID-19 on migration patterns}},
  booktitle = {Scientific},
  year = {2024},
  doi = {10.1038/s41598-024-80841-0},
}
Figure