[Paper Review] Food safety risk prediction with Deep Learning models using categorical embeddings on European Union data

Alberto Nogales, Rodrigo Díaz Morón+1 more|arXiv (Cornell University)|Sep 14, 2020

Food Supply Chain Traceability24 references4 citations

TL;DR

This paper proposes a deep learning framework using categorical embeddings with Multilayer Perceptrons and 1D Convolutional Neural Networks to predict food safety risks in the European Union. It achieves prediction accuracies between 74.08% and 93.06% for product category, hazard category, and corrective actions using EU food notification data, demonstrating the effectiveness of embedding-based models in public health risk forecasting.

ABSTRACT

The world is becoming more globalized every day and people can buy products from almost every country in the world in their local stores. Given the different food and feed safety laws from country to country, the European Union began to register in 1977 all irregularities related to traded products to ensure cross-border monitoring of information and a quick reaction when risks to public health are detected in the food chain. This information has also an enormous potential as a preventive tool, in order to warn actors involved in food safety and optimize their resources. In this paper, a set of data related to food issues was scraped and analysed with Machine Learning techniques to predict some features of future notifications, so that pre-emptive measures can be taken. The novelty of the work relies on two points: the use of categorical embeddings with Deep Learning models (Multilayer Perceptron and 1-Dimension Convolutional Neural Networks) and its application to solve the problem of predicting food issues in the European Union. The models allow several features to be predicted: product category, hazard category and finally the proper action to be taken. Results show that the system can predict these features with an accuracy ranging from 74.08% to 93.06%.

Motivation & Objective

• To develop a predictive model for food safety risks in the European Union using historical notification data.
• To investigate the effectiveness of categorical embeddings in improving deep learning performance on tabular food safety data.
• To enable early warning systems by predicting key features of future food safety notifications.
• To optimize resource allocation for food safety authorities through pre-emptive risk identification.
• To evaluate the performance of deep learning models (MLP and 1D-CNN) on multi-class classification of food safety issues.

Proposed method

• Categorical embeddings are applied to discrete features such as product type and hazard category to convert them into dense vector representations.
• A Multilayer Perceptron (MLP) and a 1D Convolutional Neural Network (1D-CNN) are trained end-to-end on the embedded features.
• The input data consists of anonymized EU food safety notifications from 1977–2019, including product category, hazard type, and corrective actions.
• Embedding layers learn hierarchical representations of categorical variables, improving generalization over raw one-hot encodings.
• The models are trained to predict three target variables: product category, hazard category, and action type.
• Model evaluation uses accuracy as the primary metric across multi-class classification tasks.

Experimental results

Research questions

• RQ1 Can categorical embeddings improve the performance of deep learning models in predicting food safety risk categories?
• RQ2 To what extent can deep learning models predict future food safety notifications based on historical EU data?
• RQ3 How do MLP and 1D-CNN architectures compare in classifying food safety features such as product and hazard categories?
• RQ4 Can the model effectively predict the corrective actions required for food safety incidents?
• RQ5 What is the predictive accuracy of the system across different food safety risk categories?

Key findings

• The proposed deep learning models achieved prediction accuracies ranging from 74.08% to 93.06% across different classification tasks.
• The 1D-CNN model outperformed the MLP in predicting hazard categories, indicating better feature extraction from sequential patterns.
• Categorical embeddings significantly improved model generalization compared to standard one-hot encoding.
• The system successfully predicted corrective actions with high accuracy, supporting operational decision-making.
• The model demonstrated strong performance in predicting product categories, with accuracy reaching 93.06%.
• The results confirm that deep learning with embeddings is a viable approach for early detection of food safety risks in large-scale EU datasets.