Note: This project was selected First Place Winner of the Health Sciences Track at Carolina Data Challenge 2024.

Carolina Data Challenge 2024 - Health Sciences Track

Team: Analysis on Influenza, Pneumonia, and COVID Trends & Travel

Welcome to our project repository for the Carolina Data Challenge 2024! We are a team competing in the Health Sciences Track, aiming to uncover insights into influenza, pneumonia, and COVID-19 trends & travel restrictions across the United States.

Project Overview

Objective

Our primary goal is to analyze and identify patterns in Total Pneumonia, Influenza, and COVID-19 Deaths across different states over time. By leveraging K-Means Time Clustering, we aim to group states with similar mortality trends, providing valuable insights into regional health dynamics.

Methodology

K-Means Time Clustering

K-Means Time Clustering is an extension of the traditional K-Means algorithm tailored for time series data. Unlike standard K-Means, which clusters based on spatial proximity, K-Means Time Clustering accounts for the temporal dimension, allowing us to group states based on the similarity of their death trends over time.

Why K-Means Time Clustering?

• Suitability for Time Series Data: Time series data involves sequences of data points indexed in time order. Traditional clustering methods may overlook the temporal dependencies and patterns inherent in such data. K-Means Time Clustering addresses this by considering the entire temporal trajectory of each state, ensuring that states with similar trends are grouped together.
• Scalability: K-Means is computationally efficient and scalable to large datasets, making it ideal for analyzing extensive time series data across all US states.
• Interpretability: The resulting clusters are straightforward to interpret, facilitating easy identification of regional patterns and trends.

Evaluation Metric

To evaluate the effectiveness of our clustering, we utilized the Silhouette Score, which measures how similar an object is to its own cluster compared to other clusters. A higher Silhouette Score indicates better-defined clusters with clear boundaries.

Key Findings

1. Regional Clustering Trends

   • Clusters Identified:

     • Southeast, Southwest, and West Clusters: States within these regions exhibited similar trends in total deaths over time, characterized by consistent spikes and declines corresponding to pandemic waves and seasonal respiratory illnesses.
     • Midwest and Northeast Clusters: States in these regions showed distinct mortality trends, differing notably from those in the Southeast and Southwest. These differences may be attributed to varying public health responses, population densities, and socio-economic factors.

   • Geospatial Visualization:

     • A geospatial chart illustrating the clustering of states is available in the rileykmeans.ipynb notebook. This map provides a visual representation of how states are grouped based on their mortality trends.

2. Correlation with Flight Cancellations

   • Utilizing a separate dataset on flight cancellations, our analysis revealed that the start of 2020 experienced a significant spike in cancellations. This spike closely mirrored the surge in COVID-19 deaths, highlighting the impact of the initial outbreak and the implementation of strict travel restrictions globally.
   • Additionally, this period saw increases in pneumonia and influenza deaths, particularly during the winter months, which traditionally experience peaks in respiratory illnesses. The alignment of these trends underscores the interplay between public health measures, virus transmission, and seasonal factors.

Conclusion

Our analysis demonstrates the effectiveness of K-Means Time Clustering in identifying and visualizing regional patterns in health-related mortality data. By clustering states based on their temporal death trends, we uncover significant regional similarities and differences, providing a foundation for targeted public health interventions and policy-making.

The correlation between flight cancellations and COVID-19 deaths further emphasizes the interconnectedness of public health responses and mobility patterns, offering insights into how global events can influence local health outcomes.

Repository Structure

• rileykmeans.ipynb: Contains the implementation of K-Means Time Clustering, geospatial visualizations, and detailed analysis of clustering results.
• kensho.ipynb: Explores the relationship between flight cancellations and COVID-19 deaths, highlighting significant temporal correlations.

Installation

1. Clone the Repository

   git clone https://github.com/yourusername/carolina-data-challenge.git
   cd carolina-data-challenge

2. Create a Virtual Environment

   python3 -m venv venv
   source venv/bin/activate

3. Install Dependencies

   pip install -r requirements.txt

Running the Analysis

1. Launch Jupyter Notebook

   jupyter notebook

2. Open and Run Notebooks

   • Navigate to rileykmeans.ipynb to explore the clustering analysis and geospatial visualizations.
   • Open kensho.ipynb to examine the correlation between flight cancellations and COVID-19 deaths.

Acknowledgements

We would like to thank the organizers of the Carolina Data Challenge for providing the platform to showcase our analytical skills and contribute to meaningful research in the health sciences domain.

Contact

Riley Harper - riley.harper@unc.edu

Kensho Pilkey - kpilkey@unc.edu

Aidan Guenthner - aidguent@unc.edu

---