A Correlation-Aware Feature Selection and Explainable AI Framework for Avian Influenza Outbreak Prediction

Timely and accurate prediction of infectious disease outbreaks is essential for effective public health response and resource allocation. Machine learning (ML) models have shown promise in forecasting such outbreaks by uncovering complex patterns in surveillance data. However, many existing models suffer from diminished interpretability and reduced performance due to the presence of redundant or highly correlated features. This issue not only affects model generalizability but also hinders domain experts from gaining actionable insights. Furthermore, most prior work in disease prediction has overlooked the combined impact of feature correlation and post-hoc model explainability. To address this gap, this study proposes a two-phase framework that integrates correlation-aware feature selection with explainable artificial intelligence (XAI). In the first phase, statistical techniques including Pearson correlation, mutual information, and variance inflation factor (VIF) are used to identify and remove redundant features. In the second phase, models are trained on both full and refined feature sets, followed by SHAP (SHapley Additive Explanations) analysis to assess feature contributions and model transparency. The framework is evaluated using avian influenza outbreak data, where refined feature subsets consistently led to improved F1-scores, AUC-PR, accuracy, and execution time. SHAP plots confirmed that retained features were more interpretable and biologically relevant, while statistical validation via paired t-tests and effect size analysis showed that the improvements were significant (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">p</i> < 0.05). Sensitivity analysis further validated the robustness of the chosen correlation threshold. The proposed approach enhances both performance and explainability, making it a robust and generalizable solution for epidemiological modeling. This work contributes a reproducible methodology that bridges statistical feature evaluation with interpretable ML, advancing practical deployment in disease surveillance systems.