MAESTRO: Multi-modal Adaptive Estimation for Temporal Respiratory Disease Outbreak

This repository contains the official implementation of the paper: MAESTRO: Multi-modal Adaptive Estimation for Temporal Respiratory Disease Outbreak.

Timely and robust influenza incidence forecasting is critical for public health decision-making. This paper presents MAESTRO, a novel, unified framework that synergistically integrates advanced spectro-temporal modeling with multi-modal data fusion, including surveillance, web search trends, and meteorological data. By adaptively weighting heterogeneous data sources and decomposing complex time series patterns, the model achieves robust and accurate forecasts.

Model Architecture

MAESTRO processes input data through a multi-stage pipeline designed to capture temporal and spectral features robustly. A moving-average block decomposes inputs into seasonal and trend components. These streams are then processed by parallel pathways including Mamba state-space models, multi-scale temporal convolutions, and a frequency-domain module. Cross-channel attention captures inter-variable dependencies, and an adaptive fusion mechanism synthesizes the insights from these diverse analytical perspectives.

Key Contributions

1. Novel Spectro-Temporal Architecture: A unified architecture that synergistically couples series decomposition, state-space modeling (Mamba), and frequency-domain learning to address the complexities of epidemiological forecasting.
2. Effective Multi-modal Integration: A cohesive strategy that merges heterogeneous data sources—epidemiological surveillance, web search trends, and weather—through cross-channel attention and an adaptive fusion framework.
3. State-of-the-Art Performance: Demonstrated superior accuracy and robustness on a comprehensive 11-year Hong Kong influenza dataset, achieving an R² of 0.956. The entire pipeline is publicly available to ensure full reproducibility.

Performance and Results

Evaluated on over 11 years (2014-2025) of Hong Kong influenza data (excluding the COVID-19 period), MAESTRO demonstrates state-of-the-art performance.

Comparison with State-of-the-Art Models

MAESTRO's predictions show a much better fit to the ground truth compared to other SOTA models, capturing both the peaks and troughs of influenza seasons with higher fidelity.

Ablation Study

Extensive ablations confirm the significant contributions of MAESTRO's multi-modal and spectro-temporal components. The full model consistently outperforms its ablated variants.

Error Distribution

The error distribution of MAESTRO is more concentrated around zero and has thinner tails compared to other models, indicating more reliable and less biased forecasts.

Getting Started

Prerequisites

• Python 3.8+
• Docker (optional, for containerized setup)
• Key Python Libraries:
  • torch: Core deep learning framework.
  • mamba-ssm: For the Mamba state-space model implementation.
  • pandas, numpy, scikit-learn: For data manipulation and processing.
  • matplotlib, seaborn, plotly: For data visualization and plotting results.
  • statsmodels, neuralforecast, darts, gluonts: For running baseline model comparisons.

A complete list of dependencies is available in requirements.txt.

Installation

1. Clone the repository:

   git clone https://github.com/Hylouis233/MAESTRO.git
   cd MAESTRO

2. Create and activate a virtual environment (recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

3. Install dependencies:

   pip install -r requirements.txt

Using Docker

For a fully containerized setup, you can use Docker and Docker Compose:

1. Build the Docker image:

   docker build -t maestro .

   or

   sh build_docker.sh

2. Run using Docker Compose:

   docker-compose up

Citation

If you find this work useful for your research, please consider citing our paper:

@misc{liu2025maestro,
  title        = {Multi-modal Adaptive Estimation for Temporal Respiratory Disease Outbreak},
  author       = {Liu, Hong and Cen, Kerui and Chen, Yanxing and Liu, Zige and Chen, Dong and Yang, Zifeng and Hon, Chitin},
  year         = {2025},
  howpublished = {arXiv preprint},
  eprint       = {2509.08578},
  archiveprefix= {arXiv},
  primaryclass = {cs.LG},
  doi          = {10.48550/arXiv.2509.08578},
  url          = {https://arxiv.org/abs/2509.08578},
  abstract     = {Timely and robust influenza incidence forecasting is critical for public health decision-making. This paper presents MAESTRO (Multi-modal Adaptive Estimation for Temporal Respiratory Disease Outbreak), a novel, unified framework that synergistically integrates advanced spectro-temporal modeling with multi-modal data fusion, including surveillance, web search trends, and meteorological data. By adaptively weighting heterogeneous data sources and decomposing complex time series patterns, the model achieves robust and accurate forecasts. Evaluated on over 11 years of Hong Kong influenza data (excluding the COVID-19 period), MAESTRO demonstrates state-of-the-art performance, achieving a superior model fit with an R² of 0.956. Extensive ablations confirm the significant contributions of its multi-modal and spectro-temporal components. The modular and reproducible pipeline is made publicly available to facilitate deployment and extension to other regions and pathogens, presenting a powerful tool for epidemiological forecasting.},
  keywords     = {time series forecasting, multi-modal learning, influenza prediction, state-space models, Transformer, frequency-domain learning, uncertainty estimation}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.