Response to the letter by Prof. Yoshiyasu Takefuji: Enhancing methodological rigor in machine learning for food authentication

Abstract

Background

Machine learning (ML) is pivotal in food authentication, yet biases in feature importance assessments and model dependency remain critical challenges, as highlighted by Prof. Takefuji.

Scope and approach

This response addresses Prof. Takefuji's concerns by proposing advanced statistical methodologies (e.g., Spearman's correlation, permutation testing), rigorous validation frameworks, and interdisciplinary collaboration to mitigate biases. We emphasize improving the reliability, fairness, and interpretability of ML models across diverse datasets and regulatory contexts.

Key findings and conclusion

Integrating robust statistical methods with domain expertise enhances model transparency and accuracy. Recommendations include adopting ensemble modelling, cross-validation, and bias audits to ensure actionable transparency for stakeholders. These steps are vital for advancing equitable and reliable food authentication systems.