Advancing freshness classification of freshly squeezed fruit juice via integrated multivariate analysis and machine learning approaches

Jiahui Ma; Xiuli Xu; Yunhe Hong; Yun Xie; Xuesong Feng; Xiujuan Wang; Yuan Zhang; Feng Zhang

doi:10.1016/j.foodchem.2025.145249

Advancing freshness classification of freshly squeezed fruit juice via integrated multivariate analysis and machine learning approaches

Food Chem. 2025 Jun 21:491:145249. doi: 10.1016/j.foodchem.2025.145249. Online ahead of print.

Authors

Jiahui Ma¹, Xiuli Xu², Yunhe Hong², Yun Xie², Xuesong Feng³, Xiujuan Wang², Yuan Zhang³, Feng Zhang⁴

Affiliations

¹ Institute of Food Safety, Chinese Academy of Quality and Inspection & Testing, Beijing 100176, China; Key Laboratory of Food Quality and Safety, State Administration for Market Regulation, Beijing 100176, China; School of Pharmacy, China Medical University, Shenyang 110122, China.
² Institute of Food Safety, Chinese Academy of Quality and Inspection & Testing, Beijing 100176, China; Key Laboratory of Food Quality and Safety, State Administration for Market Regulation, Beijing 100176, China.
³ School of Pharmacy, China Medical University, Shenyang 110122, China.
⁴ Institute of Food Safety, Chinese Academy of Quality and Inspection & Testing, Beijing 100176, China; Key Laboratory of Food Quality and Safety, State Administration for Market Regulation, Beijing 100176, China. Electronic address: fengzhang@126.com.

PMID: 40580605
DOI: 10.1016/j.foodchem.2025.145249

Abstract

In this study, multivariate analysis (MVA) and machine learning (ML), combined with UHPLC-HRMS, were used to evaluate the freshness of apples for juice production based on the analysis of freshly squeezed apple juice. A total of eight stages of apple juice, representing varying levels of freshness, were examined. Unsupervised MVA techniques, including PCA and Spearman rank correlation heatmap, revealed biochemical changes in the apple juice composition as rotten progressed. Fresh samples (J1-J3) exhibited similar chemical profiles and clustered closely, while samples in rotten stages (J6-J8) displayed dispersion, reflecting substantial metabolic alterations. With the Linear Support Vector Machine (SVM) model achieved 100 % classification accuracy in cross-validation, and demonstrating the high classification accuracy (91.3 %) for test samples. Fifteen key biomarkers associated with rotten were identified, providing robust evidence for freshness differentiation. This study demonstrates that multivariate analysis and machine learning enhances the accuracy of apple juice freshness classification, offering a robust approach with potential applications in food quality monitoring and safety assurance.

Keywords: Apple juice; Biomarkers; Food quality control; Freshness differentiation; Machine learning; Multivariate analysis.