Interpretable spatio-temporal prediction using Deep Neural Network - Local Interpretable Model-agnostic Explanations: A case study on leptospirosis outbreaks in Malaysia

Leptospirosis is a widespread zoonotic disease with complex spatio-temporal dynamics. This study investigates the use of Deep Neural Network (DNN) in combination with Local Interpretable Model-Agnostic Explanations (LIME) for weekly spatio-temporal predictions of leptospirosis occurrence. The predictive model integrates hydroclimatic and environmental data to assess its effectiveness in predicting leptospirosis cases and quantifying key input variables in Negeri Sembilan, Malaysia. Using a DNN architecture with hyperparameter tuning via grid search, we developed a globally trained model that achieved an overall prediction accuracy of 70.5% across 214 pixels. We identified acidic soil and a higher presence of rubber plantations as strong predictors of leptospirosis occurrence. Additionally, mean temperature and minimum rainfall emerged as important hydroclimatic contributors. These insights enable public health authorities to proactively identify and prioritize high-risk areas for targeted interventions, improving disease mitigation strategies. Furthermore, the methodology is adaptable to other regions with similar environmental and socio-economic conditions, strengthening early warning systems and enhancing preparedness against future leptospirosis outbreaks. While demonstrated on leptospirosis prediction, the proposed DNN-LIME framework is adaptable to spatio-temporal challenges in diverse domains such as supply chain optimization, urban planning, and industrial risk management. The integration of interpretability via LIME ensures actionable insights for stakeholders beyond public health, bridging the gap between complex models and real-world decision-making.

Text 121152.pdf Restricted to Repository staff only Download (9MB) Official URL or Download Paper: https://linkinghub.elsevier.com/retrieve/pii/S0952...

Actions (login required)

View Item