This repository contains the code and results for the MSc thesis:
Online Correlation Changepoint Detection via Adaptive EWMA Charts
Jilly Lim — MSc Machine Learning and Data Science (Online), Imperial College London, 2025
This project investigates real-time detection of changes in correlation between two time series, using rolling Pearson correlation as a test statistic. We propose an Adaptive EWMA control chart and compare it against three established methods:
- CUSUM – classical mean-shift detector
- Wilcoxon Rank-Sum Scan – nonparametric windowed test
- MEWMA – multivariate EWMA applied to bivariate raw stream
Both simulated studies and real-world case studies are included:
- Financial data (e.g., NVDA vs GOOGL, Zoom vs S&P 500)
- Public health data (COVID-19 case rates: Los Angeles vs San Francisco)
| File | Description |
|---|---|
thesis_clean.ipynb |
Full code (no output, minimal size) |
thesis_clean.html |
HTML version of above |
thesis_results.ipynb |
Full code with all outputs |
thesis_results.html |
HTML version with outputs |
Both notebooks contain identical code and analysis; only the presence of outputs differs.
- Code is written in Python 3 and intended for reproducibility.
- Datasets are publicly available via Kaggle and are downloaded using
kagglehub. - Some sections (e.g., simulation runs) may take several minutes to execute.
This project uses standard Python packages including:
numpy,pandas,matplotlib,scipykagglehub(for downloading public datasets)
Install with:
pip install numpy pandas matplotlib scipy kagglehub