Anticipating viruses with sewage water data
KickstartAI - AI predictive model
While the worst seems to be behind us, the risk of a virus outbreak remains. That's why a team of data scientists and engineers from Ahold Delhaize, ING, KLM, Philips and NS joined forces during the 2020 crisis. We developed a model that uses viral load data from wastewater systems to predict outbreaks before they spread.
Reducing virus impact with AI-powered forecasting
We developed an early warning system for virus-related hospital inflow. The aim was to provide decision-makers with real-time insights to better anticipate hospital capacity.
Helping them gain a more proactive response in sudden surge and make optimal use of hospital resources and prevent code black situations. To achieve this we have outlined the following subgoals:
Develop an AI-powered early warning system
Improve forecasting accuracy
Integrate AI predictions into decision-making tools
Enable proactive hospital capacity planning
Bridge AI research and real-world applications
Approach
KickstartAI organised a sprint-based approach, uniting data scientists from NS, ING, and Ahold Delhaize. Over three sprints, we developed, tested, and refined the predictive model. Using gradient-boosted trees and time series forecasting, we analyzed hospital data, environmental factors, and wastewater trends. The model was integrated into the dashboard, providing policymakers with actionable insights.
Sprint 1: Data exploration and baseline model development.
Sprint 2: Model iterations, feature engineering, and initial testing.
Sprint 3: Refining predictions and integrating results into operational dashboards.
Through rigorous data preprocessing, the team developed a model capable of outperforming naive forecasting techniques, significantly enhancing predictive accuracy.
Key features
Early detection from wastewater measurements detected case surges at the earliest stage.
Using regional mobility, environmental data, and past hospitalization trends for accuracy.
The model is scalable and adaptable, ensuring flexibility for future applications.
Helping the government anticipate for future outbreaks
This project proves how AI can enhance healthcare by providing critical insights. The AI-driven forecasting model is now in real-world testing, with improvements based on feedback. This project has set a foundation for using AI in public health challenges. It’s a great showcase of how cross-industry collaboration can solve societal issues.
Soon available to the government, this platform-independent early warning system helps anticipate outbreaks and protect public health.
Results
Improved short-term forecasting accuracy by 26% over baseline models.
Enhanced early warning system for COVID hospital admissions.
Seamless integration into existing dashboards used by government decision-makers.
Sustainable AI deployment, with a production-ready model.