Explainable Artificial Intelligence techniques for interpretation of food datasets: a review

TL;DR

This review maps explainable AI (XAI) applications in food datasets, proposing a taxonomy that links data types (tabular, pictorial, spectral, time series) with explanation forms (numerical, visual, rule-based, textual, mixed) to food quality tasks. It surveys XAI use across food safety, nutritional value, sensory characteristics, authenticity/traceability, and sustainability/healthiness, highlighting CAM-based visual explanations for images, SHAP/LIME for numeric attributions, and limited use of rule-based approaches. Key contributions include a comprehensive classification of data types and explanation forms, a synthesis of current trends, and a discussion of open challenges such as underexplored spectral and time-series data, and the need for glocal explanations and counterfactuals. The findings underscore significant opportunities to advance trustworthy AI in the food sector by promoting standardized terminology, multi-modal data fusion explainability, and practical, end-user focused explanations that bridge AI outputs with domain expertise.

Abstract

Artificial Intelligence (AI) has become essential for analyzing complex data and solving highly-challenging tasks. It is being applied across numerous disciplines beyond computer science, including Food Engineering, where there is a growing demand for accurate and trustworthy predictions to meet stringent food quality standards. However, this requires increasingly complex AI models, raising reliability concerns. In response, eXplainable AI (XAI) has emerged to provide insights into AI decision-making, aiding model interpretation by developers and users. Nevertheless, XAI remains underutilized in Food Engineering, limiting model reliability. For instance, in food quality control, AI models using spectral imaging can detect contaminants or assess freshness levels, but their opaque decision-making process hinders adoption. XAI techniques such as SHAP (Shapley Additive Explanations) and Grad-CAM (Gradient-weighted Class Activation Mapping) can pinpoint which spectral wavelengths or image regions contribute most to a prediction, enhancing transparency and aiding quality control inspectors in verifying AI-generated assessments. This survey presents a taxonomy for classifying food quality research using XAI techniques, organized by data types and explanation methods, to guide researchers in choosing suitable approaches. We also highlight trends, challenges, and opportunities to encourage the adoption of XAI in Food Engineering.

Figures (7)

• Figure 1: Overview scheme, from food quality tasks to XAI techniques. XAI is applied as an endpoint of a data processing pipeline that takes into consideration the task, type of data, and the specific AI model employed, e.g., Machine Learning and Deep Learning. According to these factors, one or more specific XAI techniques are employed, which produce explanations---tokens of information useful for model developers or users to gain insights into the prediction dynamics. Explanations can be produced in different types, each conveying a different facet of the information provided.
• Figure 2: Chart illustrating approximately the trade-off between expressivity or flexibility and interpretability. Expressive models, such as those based on DL, are capable of reaching higher task-level performance but are often hardly interpretable. On the other hand, less complex models, like LR, are inherently interpretable, but often incapable of attaining high task-level performance.
• Figure 3: Representation of the types of explanations provided by XAI techniques, along with a summary of their key advantages.
• Figure 4: Distribution of papers surveyed in the present work per publication year and data type. Most of the articles were published from 2020 onwards, illustrating how the application of XAI to food quality topics is a rather new and evolving discipline.
• Figure 5: Distribution of papers surveyed in the present work per topic and explanation type.