Cross-Correlation and Causality Analysis for Wine Quality: A Stacked Machine Learning Approach.

Abstract

Accurately revealing the complex relationship between the physicochemical properties of wine and its quality is of significant importance for optimizing brewing processes and improving the quality of low-quality wines. This study employs multifractal detrended cross-correlation analysis (MF-DCCA) and transfer entropy (TE) methods to systematically explore the MF cross-correlation and causal mechanisms between 11 physicochemical properties and wine quality. The results reveal that volatile acidity, sulfates, and residual sugar in red wine, as well as fixed acidity, citric acid, and volatile acidity in white wine, exhibit pronounced MF features with wine quality, as identified by MF-DCCA. Furthermore, TE analysis confirms causal relationships between these variables and wine quality, highlighting their critical influence on quality formation. On the basis of these findings, we constructed a stacking ensemble machine learning model that integrates multiple base models to enhance the accuracy and generalization ability of wine quality prediction. Experimental results demonstrate that the proposed model not only significantly outperforms conventional single algorithms but also enables effective enhancement of low-quality wine samples through targeted adjustment of specific physicochemical properties. This study provides an efficient data-driven framework for wine quality assessment and optimization and offers theoretical support for precision quality control and process improvement.