Explainable artificial intelligence for predicting dengue outbreaks in Bangladesh using eco-climatic triggers

Abstract

Background

Dengue represents a significant public health threat in Bangladesh, characterized by its complex ecological transmission dynamics. To improve dengue prevention and control efforts, firstly, we employ state-of-the-art artificial intelligence (AI) methods to identify the roles of eco-climatic factors in predicting dengue outbreaks in Bangladesh.

Methods

We utilize high-performance machine learning (ML) models, XGBoost and LightGBM, combined with explainable AI (XAI) methodologies to evaluate the predictive performance and impact of various dengue determinants in Bangladesh from 2000 to 2023. The LightGBM and XGBoost models were also utilized to predict dengue cases and early warning trends from 2024 to 2030. Climatic, socio-demographic, and landscape features were used to train these models; SHapley Additive Explanations (SHAP) values and LIME (Local Interpretable Model-agnostic Explanations) were used to interpret the results.

Findings

Between 2000 and 2023, Bangladesh experienced the highest number of dengue cases in August, while November saw the most fatalities. The XGBoost model excelled in predicting dengue outbreaks, achieving an AUC score of 0.89, a Log Loss of 0.64. Key predictors identified by the model include population density, precipitation, temperature, and land-use patterns. Additionally, Local Interpretable Model-agnostic Explanations (LIME) provided insights into the model's predictions, highlighting the significance of population density, relative humidity, and minimum temperature in dengue outbreaks.

Interpretation

This study showcases the potential of XAI in uncovering the complexities of dengue outbreaks, providing a robust tool for public health interventions. Our AI-driven framework can be utilized to generate prompt and timely alerts to prevent imminent dengue and other infectious disease outbreaks.