COVID-19 Predictor

Predictive modeling of COVID-19 hospital admissions and deaths in the UK using Google mobility data as predictor variables.

Overview

This project implements a rolling-window time-series forecasting approach to predict:

• Hospital admissions (1-4 weeks ahead)
• Deaths (1-4 weeks ahead)

Using publicly available data sources:

• Google COVID-19 Community Mobility Reports (predictors)
• UK Health Security Agency COVID-19 data (targets)

Key Features

• Temporal data preprocessing: Interpolation and 14-day smoothing to handle reporting artifacts
• Rolling window forecasting: Fixed training windows with linear regression
• Training window sensitivity analysis: Comparison of 2, 4, 8, and 16-week windows
• Visual lag analysis: Identification of 2-3 week mobility→hospitalization lag

Results

• Hospital Cases (4-week ahead): R² = 0.858 (8-week training window)
• Deaths (4-week ahead): Lower accuracy due to increased downstream variability

Limitations

⚠️ No train/test split: All metrics computed on overlapping data; R² values likely overestimate true performance

Requirements

pandas
numpy
scikit-learn
matplotlib
seaborn

Data Sources

• Google COVID-19 Mobility Data
• UK COVID-19 Archive (not included; 1.6GB)

Usage

1. Download UK COVID-19 archive and extract to covid-19-archive/
2. Run all cells in Covid-predictor.ipynb

License

Educational project for Applied Machine Learning coursework.