Influenza Modeling Based on Massive Feature Engineering and International Flow Deconvolution

Ziming Liu, Yixuan Wang, Zizhao Han, Dian Wu

Introduction

Predicting the global spread and mortality impact of influenza remains a critical challenge for public health. Traditional epidemiological models, such as the Susceptible-Infectious-Recovered (SIR) framework, primarily focus on local infection mechanics. However, they often fail to integrate the complex web of socio-economic and environmental factors that drive mortality across diverse nations. Furthermore, in an interconnected world, a country's disease burden is significantly influenced by international flows of people and goods. Researchers from Peking University have proposed a novel framework that addresses these gaps by combining deep learning-based feature extraction with a "deconvolution" approach to isolate local drivers from global influences.

Methodology

The proposed framework utilizes a multi-stage pipeline designed to handle high-dimensional, sparse data and account for international connectivity.

Overview of the multi-stage modeling pipeline, integrating data completion, feature extraction, and international flow deconvolution.

Data Preparation and Periodicity Analysis

The study began with 1,170 socio-economic indicators from the World Bank, covering economy, health, sanitation, and environment for 183 countries. To address missing values, the researchers employed Matrix Completion, ensuring a dense dataset. Discrete Fourier Transform (DFT) analysis of influenza activity (from FluNet/GISRS) confirmed a strict one-year global periodicity, justifying the use of annual data for the model.

Hybrid Feature Extraction

To manage the high dimensionality of the indicators, the authors developed a hybrid pipeline. A 4-layer Autoencoder (AE) was used to compress the 1,170 dimensions into a 10-dimensional non-linear manifold. Subsequently, Principal Component Analysis (PCA) was applied to these dimensions to extract six orthogonal principal characteristics. This "Kernel PCA-like" approach effectively captures non-linear relationships while maintaining the interpretability of linear components.

International Flow Deconvolution

To account for international "spillover" effects, the researchers introduced a higher-order rectification term. The model treats international migration and trade as a convolution kernel ($M$) acting on regional activity ($z$). By solving this deconvolution problem ($\epsilon Mz$), the framework isolates purely regional mortality factors from those introduced by global connectivity.

Results and Discussion

The effectiveness of the data preprocessing was validated using a Support Vector Machine (SVM) to classify countries as "Developed" or "Developing." While the SVM failed on raw, incomplete data, it achieved high accuracy on the "completed" dataset, confirming the robustness of the matrix completion step.

SVM classification results demonstrating the clear separation of developed and developing countries after data completion.

Model Performance and Stability

Bootstrap cross-validation indicated that the "rectified" model, which includes international flows, was significantly more stable and accurate than a simple regional linear model. The AE-based feature extraction also outperformed standard PCA in terms of reconstruction error, suggesting that the underlying drivers of influenza mortality are inherently non-linear.

Socio-Economic Insights

The model identified key drivers of mortality. Factors positively correlated with higher mortality ("Bad" features) included greenhouse gas emissions and rural poverty gaps. Conversely, "Good" features associated with lower mortality included the density of community health workers and average precipitation levels. The model also provided region-specific insights; for example, environmental factors were primary drivers in Nigeria, whereas community healthcare was the most significant lever in Japan.

Conclusion

By integrating massive feature engineering with international flow deconvolution, this framework provides a robust tool for understanding global influenza mortality. It successfully decouples local policy impacts from the noise of international travel. Future work will aim to increase temporal resolution and model migration as a dynamic Markov process to further refine predictive accuracy and policy targeting.

Summary updated

Abstract

In this article, we focus on the analysis of the potential factors driving the spread of influenza, and possible policies to mitigate the adverse effects of the disease. To be precise, we first invoke discrete Fourier transform (DFT) to conclude a yearly periodic regional structure in the influenza activity, thus safely restricting ourselves to the analysis of the yearly influenza behavior. Then we collect a massive number of possible region-wise indicators contributing to the influenza mortality, such as consumption, immunization, sanitation, water quality, and other indicators from external data, with $1170$ dimensions in total. We extract significant features from the high dimensional indicators using a combination of data analysis techniques, including matrix completion, support vector machines (SVM), autoencoders, and principal component analysis (PCA). Furthermore, we model the international flow of migration and trade as a convolution on regional influenza activity, and solve the deconvolution problem as higher-order perturbations to the linear regression, thus separating regional and international factors related to the influenza mortality. Finally, both the original model and the perturbed model are tested on regional examples, as validations of our models. Pertaining to the policy, we make a proposal based on the connectivity data along with the previously extracted significant features to alleviate the impact of influenza, as well as efficiently propagate and carry out the policies. We conclude that environmental features and economic features are of significance to the influenza mortality. The model can be easily adapted to model other types of infectious diseases.
BibTeX
@misc{liu2019influenza,
  author = {Ziming Liu and Yixuan Wang and Zizhao Han and Dian Wu},
  title = {{Influenza Modeling Based on Massive Feature Engineering and International Flow Deconvolution}},
  year = {2019},
  eprint = {1912.02989},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1912.02989},
}
Figure