In this study, we propose an Epidemic-guided deep learning (EGDL) framework by integrating the epidemic knowledge of a modified Networked SIR model with data-centric deep learners for spatiotemporal forecasting of epidemics. Our study focuses on publicly available Tuberculosis (TB) incidence cases recorded in different prefectures of Japan[1] and various provinces of mainland China[3].
Usage of the repository for the paper "Epidemic-guided deep learning for spatiotemporal forecasting of Tuberculosis outbreak [2]".
- We introduce a modified networked SIR model by considering the saturated incidence rate and graph Laplacian diffusion to model the spatial transmission dynamics of TB outbreaks. The key epidemiological principles including positivity and boundedness are verified. The global stability analysis of the modified networked SIR model is performed using Green's formula and comparison principle for both disease-free and endemic equilibria.
- We develop two EGDL architectures namely, EGDL-Parallel and EGDL-Series frameworks by integrating the epidemic principles of the networked SIR model with data-driven deep learners. In the EGDL-Parallel framework, we provide a hybrid input of historical TB surveillance data and estimated infected curve of the networked SIR model to predict the future spatiotemporal dynamics of TB incidences. On the other hand, the EGDL-Series approach adopts a residual remodeling setup, where the stochastic variations in the disease incidences are modeled with deep learners and the deterministic counterparts are estimated by the networked SIR model.
-
In this repository, we present the application of the proposed EGDL model with Japan TB datasets from 47 prefectures [1].
-
The EGDL Code file contains the source code and the implementation of the proposed EGDL model.
-
The forecasts generated by the EGDL models along with the data-centric deep learners and the estimated infected curve of networked SIR are visually presented as follows
-
Results obtained in the paper for the epidemic datasets can directly be computed using the implementation files in this repository for the sake of replicability and reproducibility of our paper.
Barman, M., Panja, M., Mishra, N., & Chakraborty, T. (2025).
Epidemic-guided deep learning for spatiotemporal forecasting of Tuberculosis outbreak.
arXiv preprint arXiv:2502.10786.
@article{barman2025epidemic,
title={Epidemic-guided deep learning for spatiotemporal forecasting of Tuberculosis outbreak},
author={Barman, Madhab and Panja, Madhurima and Mishra, Nachiketa and Chakraborty, Tanujit},
journal={arXiv preprint arXiv:2502.10786},
year={2025}
}