International Joint Research Paper with Burapha University in Thailand Published in IEEE Access — Proposing a New Method for Analyzing Infectious Disease Surveillance Data Using Explainable AI —

Professor Takafumi Nakanishi of the TransMedia Tech Lab, Faculty of Computer Science at our university, in collaboration with Mr. Wittama Thumcharoen, Dr. Watcharaphong Yookwan, Associate Professor Krisana Chinnasarn, and Dr. Athita Onuean, has published a research paper titled “Top-K Pandemic feature selection using Approximate Inverse Model Explanation,” which has been published in the international academic journal IEEE Access.

This study proposes the “Top-K Pandemic Feature Selection” framework, which uses Approximate Inverse Model Explanation (AIME)—a method of explainable AI—to extract key features related to infectious disease outbreaks in a time-series context.

AIME is a model-agnostic explainable AI method previously proposed by Professor Nakanishi. It is characterized by its ability to estimate the importance of features from observed data without relying on the internal structure of the predictive model. In this study, we applied AIME to infectious disease surveillance data to analyze the time-varying important factors for multiple infectious diseases in Thailand, Japan, and South Korea.

In conventional infectious disease forecasting, the focus has been on predicting future case numbers using machine learning models. However, for public health decision-making, it is also crucial to understand, in an interpretable manner, which temporal, regional, and seasonal factors are associated with fluctuations in case numbers. A key feature of this study is that, rather than training a prediction model and then explaining its results, it estimates Global Feature Importance (GFI) from weekly surveillance data to analyze the factors that change over time.

The analysis utilized weekly infectious disease surveillance data from Thailand, Japan, and South Korea, comparing five diseases: dengue fever, chikungunya, measles, COVID-19, and hand, foot, and mouth disease (HFMD). We standardized data from each country into a common analytical framework and extracted time-varying important features using a 26-week rolling window, incorporating lag features, moving average features, seasonality, disease ratios, and regional information.

The results showed that lag features from weeks 1 to 4 and moving average features from weeks 4 to 12 were significant for many countries and diseases. This indicates that recent infection trends provide crucial clues for understanding future epidemic dynamics. We also confirmed that features reflecting seasonality, urban/regional characteristics, and inter-disease correlations vary in importance depending on the country and disease.

Furthermore, validation using the top 10 features selected by AIME demonstrated that predictive performance could be maintained or improved compared to using the full set of features or an autoregressive model. Consequently, the proposed method is expected to serve as an effective analytical framework for organizing infectious disease surveillance data in a more interpretable manner and for comparing epidemic factors across countries and diseases.

These results stem from an international collaborative research project applying explainable AI to infectious disease surveillance, and future applications are anticipated in data analysis, infectious disease monitoring, and early warning support within the public health sector.

■Future Plans
The Faculty of Computer Science and the TransMedia Tech Lab at our university will continue to advance research on explainable AI, data science, infectious disease surveillance, and AI-based public health support, while further developing international collaborative research with overseas research institutions, including Burapha University.

■Published Research Findings
Paper Title:
Top-K Pandemic Feature Selection using Approximate Inverse Model Explanation

Authors:
Wittama Thumcharoen, Watcharaphong Yookwan, Takafumi Nakanishi, Krisana Chinnasarn and Athita Onuean

Journal:
IEEE Access

DOI:
10.1109/ACCESS.2026.3697395

IEEE Xplore:
https://ieeexplore.ieee.org/document/11536075

Affiliation:
Faculty of Informatics, Burapha University, Thailand
School of Computer Science, Tokyo University of Technology, Japan

Related Links:
Feature article by the Faculty of Informatics, Burapha University
https://www.informatics.buu.ac.th/2020/?p=15555