Machine Learning Prediction Project

Abstract

Among the pandemics officially declared by the WHO, H1N1 and COVID-19, which both emerged after the 20th century, lasted for more than a year and had a significant impact on daily life, as well as on social, political and economic systems. In particular, the COVID-19 pandemic remains ongoing.

This study analyses changes in the three major U.S. stock market indices during the H1N1 and COVID-19 pandemics and forecasts stock price fluctuations for the following year. We clearly identified a relationship between pandemics and stock prices. This relationship was modelled using a Graph Attention Network (GAT), and the number of COVID-19 infections for a one-year period was predicted using the Exponential Smoothing State Space Model (ETS). In addition, the Prophet forecasting model was used to predict the closing prices of each index for the upcoming year, based on variables including the number of COVID-19 cases predicted by the ETS model and the infection figures recorded during the H1N1 pandemic.

Results & Metrics

• Metric A: 0.97, 0.96, 0.96
• RMSE: 890, 330, 140

Tech Stack

Python, Graph Attention Network (GAT), Exponential Smoothing State Space Model (ETS), Prophet.

Code

View on GitHub

Paper (Korean)

Original paper (Korean) is available for download

Presentation (PDF)

You can check the project presentation here — Korean PDF available below.

Download