A machine learning approach: Seasonal impact of climate change on Vibrio contamination in food products in the Dutch market

Global climate change is predicted to alter precipitation and temperature patterns worldwide, affecting a range of infectious diseases and particularly foodborne infections from Vibrio pathogens. Vibriosis is a serious illness caused by an infection with Vibrio bacteria. Infections caused by non-cholera Vibrio spp. have expanded globally over the last decades, with continued increases expected in the future. The emergence of Vibrio pathogens associated with human illnesses in Europe is linked to climate change, namely the increase in global sea surface temperature that is affecting our oceans. Recent heatwaves in northern Europe in 2018 and 2019 have also shed light on non-travel-related vibriosis as related to seasonality; an increase in future associated Vibrio spp. human infections are projected. Given predicted global climate changes, also as related to the burden of human infections with vibriosis, it is crucial to be able integrate data on regional climate change and adapt, when needed, current food monitoring programmes in the Netherlands.

Project description

Adaptation to climate change is a way of responding to global climate change. According to The Intergovernmental Panel on Climate Change (IPCC), adaptation is a sequence of modifications to actual or expected climate and the impacts of the modification. One of the climate change adaptation strategies related to food security/safety is integrating climate change concerns into existing planning and monitoring approaches. Monitoring and predicting the seasonality of microbial contaminants such as Vibrio assist food safety authorities and food business operators (FBOs) to be better prepared for climate change challenges on food safety in the Netherlands. Understanding the connection between climate change and human health helps to implement countermeasures to minimise food safety risks. As such, the modelling approach contributes to evidence-based solutions to adapt and integrate climate change concerns into current food safety monitoring programmes. Furthermore, the outcome can be used as input for policy-making related to human health risks, for example, in drafting information campaigns informing consumers about increased risks related to climate change.

In this project, an interdisciplinary approach is applied, connecting ML, climate knowledge, and food safety expertise. Data from Dutch food monitoring on Vibrio will be combined with climate data, allowing us to model the link between climate (change) and Vibrio contamination in food. Modelling will be done using several ML algorithms that have shown recent success in the domain of tabular data, like boosting algorithms (e.g. XGBoost or CatBoost), neural networks for (small) tabular data (e.g. SAINT) and classic ML models (e.g. logistic regression or a random forest). In order to combat the rather severe imbalance between contaminated and uncontaminated samples, we will implement techniques to overcome this imbalance, like under-sampling, over-sampling and class weighting.

Deliverables

• Quantitative results of Vibrio increase per degree increase for the Netherlands in 2050
• Processed data for machine learning algorithms. Figure 1 shows a summary of Vibrio data for algae and shellfish from the NVWA monitoring program from 2019 to 2023.
• Proof of concept model