AutoML-Powered Adaptive Drift Detection for COVID-19 Forecasting

This project implements an AutoML-driven adaptive drift detection framework for COVID-19 forecasting, leveraging ADWIN & DDM to dynamically adjust models in response to concept drift. It enhances predictive accuracy by intelligently retraining models, ensuring robust time-series forecasting under evolving data conditions.

Overview

This repository contains the implementation of an AutoML-driven adaptive drift detection framework for COVID-19 forecasting. The project leverages AutoSklearn, ADWIN, and DDM to dynamically adjust models in response to concept drift, ensuring robust time-series forecasting under evolving data conditions.

Task / Problem Statement

Main Problem:

The dynamic nature of COVID-19 data introduces concept drift, reducing the accuracy of static forecasting models. This project aims to enhance forecasting robustness by detecting and adapting to these shifts in real time.

Objectives:

1. Develop a baseline forecasting model using AutoML to predict COVID-19 cases and deaths.
2. Implement an adaptive forecasting model that dynamically detects and reacts to drift.
3. Compare the performance of baseline and adaptive models.

Goals & Sub-Goals

Main Goals:

• Improve COVID-19 forecasting accuracy.
• Reduce the impact of concept drift on predictions.

Sub-Goals:

• Integrate ADWIN and DDM for drift detection.
• Automate model retraining upon drift detection.
• Provide comparative performance metrics and visualizations.

Implementation Details

Code Structure:

1. Data Preprocessing and Visualization.
2. Baseline Model Training using Auto-Sklearn.
3. Drift Detection Mechanism (ADWIN, DDM).
4. Baseline Model Evaluation.
5. Adaptive Model Training with Drift Response.
6. Adaptive Model Evaluation.
7. Comparative Analysis between baseline and adaptive models.

Algorithms & Mathematical Formulations:

1. AutoML (AutoSklearn): Automates hyperparameter tuning and model selection.
2. Drift Detection:
   • ADWIN (Adaptive Windowing) detects drift by monitoring distribution changes.
   • DDM (Drift Detection Method) tracks error rates for drift detection.

Equations Used:

• Mean Absolute Error (MAE): [ MAE = \frac{1}{n} \sum_{t=1}^{n} |y_t - \hat{y_t}| ]
• Root Mean Squared Error (RMSE): [ RMSE = \sqrt{\frac{1}{n} \sum_{t=1}^{n} (y_t - \hat{y_t})^2} ]
• R-squared (R²): [ R^2 = 1 - \frac{\sum_{t=1}^{n} (y_t - \hat{y_t})^2}{\sum_{t=1}^{n} (y_t - \bar{y})^2} ]

Results & Metrics

Baseline Model Performance:

Metric	Cases	Deaths
RMSE	4760.92	99.91
MAE	3781.14	92.05
R²	-4.56	-8.89

Adaptive Model Performance:

Metric	Cases	Deaths
RMSE	3128.65	94.99
MAE	1775.07	61.38
R²	0.27	-0.63

Drift Detection Results:

Drift Index	Model	Detection Method
575	Deaths	ADWIN
607	Deaths	ADWIN
639	Deaths	ADWIN

Visualizations

Baseline Model Predictions:

Adaptive Model Predictions:

How to Use

1. Clone the repository:

   git clone https://github.com/your-repo-name.git
   cd your-repo-name

2. Install dependencies:

   pip install -r requirements.txt

3. Run the Jupyter notebooks in sequence:

   jupyter notebook

4. View results and metrics in the outputs/ directory.

Future Work

• Experiment with additional drift detection algorithms.
• Optimize AutoML for larger datasets.
• Incorporate real-time data streams.

References

• AutoSklearn Documentation
• ADWIN - Adaptive Windowing
• DDM - Drift Detection Method
• COVID-19 Forecasting Methodologies
• Auto-ML COVID-19 Forecasting Research Paper