Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MiCA: A Mobility-Informed Causal Adapter for Lightweight Epidemic Forecasting

Suhan Guo, Jiahong Deng, Furao Shen

TL;DR

MiCA addresses the challenge of incorporating mobility into epidemic forecasting under noisy, data-scarce conditions by learning a causal mobility prior from historical mobility data and injecting it into lightweight temporal forecasters via gated residual mixing. The approach avoids heavy graph networks or full attention by using an architecture-agnostic module that gates the prior with global and edge-level controls and enforces leakage-free integration. Empirical validation on four real-world datasets (COVID-19 incidence, COVID-19 mortality, influenza, dengue) shows consistent improvements for RAM and DLinear backbones, achieving average relative error reductions around 5-9% and competitive performance with state-of-the-art spatio-temporal models at a fraction of the computational cost. The work highlights the practicality of exploiting causal mobility priors as soft structure to enhance epidemic forecasting in realistic, noisy settings.

Abstract

Accurate forecasting of infectious disease dynamics is critical for public health planning and intervention. Human mobility plays a central role in shaping the spatial spread of epidemics, but mobility data are noisy, indirect, and difficult to integrate reliably with disease records. Meanwhile, epidemic case time series are typically short and reported at coarse temporal resolution. These conditions limit the effectiveness of parameter-heavy mobility-aware forecasters that rely on clean and abundant data. In this work, we propose the Mobility-Informed Causal Adapter (MiCA), a lightweight and architecture-agnostic module for epidemic forecasting. MiCA infers mobility relations through causal discovery and integrates them into temporal forecasting models via gated residual mixing. This design allows lightweight forecasters to selectively exploit mobility-derived spatial structure while remaining robust under noisy and data-limited conditions, without introducing heavy relational components such as graph neural networks or full attention. Extensive experiments on four real-world epidemic datasets, including COVID-19 incidence, COVID-19 mortality, influenza, and dengue, show that MiCA consistently improves lightweight temporal backbones, achieving an average relative error reduction of 7.5\% across forecasting horizons. Moreover, MiCA attains performance competitive with SOTA spatio-temporal models while remaining lightweight.

MiCA: A Mobility-Informed Causal Adapter for Lightweight Epidemic Forecasting

TL;DR

MiCA addresses the challenge of incorporating mobility into epidemic forecasting under noisy, data-scarce conditions by learning a causal mobility prior from historical mobility data and injecting it into lightweight temporal forecasters via gated residual mixing. The approach avoids heavy graph networks or full attention by using an architecture-agnostic module that gates the prior with global and edge-level controls and enforces leakage-free integration. Empirical validation on four real-world datasets (COVID-19 incidence, COVID-19 mortality, influenza, dengue) shows consistent improvements for RAM and DLinear backbones, achieving average relative error reductions around 5-9% and competitive performance with state-of-the-art spatio-temporal models at a fraction of the computational cost. The work highlights the practicality of exploiting causal mobility priors as soft structure to enhance epidemic forecasting in realistic, noisy settings.

Abstract

Accurate forecasting of infectious disease dynamics is critical for public health planning and intervention. Human mobility plays a central role in shaping the spatial spread of epidemics, but mobility data are noisy, indirect, and difficult to integrate reliably with disease records. Meanwhile, epidemic case time series are typically short and reported at coarse temporal resolution. These conditions limit the effectiveness of parameter-heavy mobility-aware forecasters that rely on clean and abundant data. In this work, we propose the Mobility-Informed Causal Adapter (MiCA), a lightweight and architecture-agnostic module for epidemic forecasting. MiCA infers mobility relations through causal discovery and integrates them into temporal forecasting models via gated residual mixing. This design allows lightweight forecasters to selectively exploit mobility-derived spatial structure while remaining robust under noisy and data-limited conditions, without introducing heavy relational components such as graph neural networks or full attention. Extensive experiments on four real-world epidemic datasets, including COVID-19 incidence, COVID-19 mortality, influenza, and dengue, show that MiCA consistently improves lightweight temporal backbones, achieving an average relative error reduction of 7.5\% across forecasting horizons. Moreover, MiCA attains performance competitive with SOTA spatio-temporal models while remaining lightweight.
Paper Structure (24 sections, 1 theorem, 33 equations, 5 figures, 2 tables)

This paper contains 24 sections, 1 theorem, 33 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Spatio-Temporal Forecasting
  4. Causality-Aware Time-Series Modeling
  5. Methods
  6. Preliminary
  7. Transformer-based Temporal Framework
  8. Multi-Head Attention (MHA)
  9. Full Attention Block
  10. RAM-Pruned Temporal Backbone
  11. Mobility-Causal Graph Discovery
  12. Mobility-Informed Causal Adapter (MiCA)
  13. Causal Residual Mixer (CRM)
  14. Prior-Gated Projection (PGP)
  15. Adpater Loss Function
  16. ...and 9 more sections

Key Result

Theorem 1

Let $\mathcal{F}_\theta$ denote the forecasting network after the MiCA injection point, viewed as a mapping from the fused representation to the output. Assume $\mathcal{F}_\theta$ is $L$-Lipschitz under the Frobenius norm. For a MiCA update of the form $\tilde{Z} = Z + W_o(\lambda \, \bar{S} Z)$, w In particular, the influence of MiCA is explicitly controlled by the global mixing weight $\lambda$

Figures (5)

  • Figure 1: Modeling challenges under real-world epidemic data.
  • Figure 2: End-to-end MiCA pipeline. PCMCI derives causal mobility relations, which MiCA incorporates using link-level gating (PGP) and gated residual mixing (CRM). The model fuses mobility-aware spatial features with a RAM-pruned temporal encoder, trained under prediction and prior-regularization losses.
  • Figure 3: Ablation analysis on major components of the proposed model.
  • Figure 4: RMSE comparison across forecasting horizons (7–42 steps) on COVID-19 mortality.
  • Figure 5: Ablation analysis on causal discovery of the proposed model.

Theorems & Definitions (1)

  • Theorem 1: Bounded Influence of MiCA