A Multidimensional Feature Optimization and Machine Learning-Based Predictive Framework for Automated Wine Quality Assessment

Abstract Assessment of wine quality is an essential step in the wine industry, where traditional evaluation methods mainly depend on expert sensory analysis and physicochemical testing. However, these conventional approaches are often time-consuming, expensive, and influenced by subjective human judgment. In this context, wine quality evaluation by alternative computer based-methods is gaining substantial interest to overcome traditional assessment related challenges. In this context, the present study aimed at investigating the wine quality by machine learning based approaches using available wine datasets of different origin and to promote the speed of automation in wine industry. The proposed framework integrates both feature selection methods (e.g., multiple dimensionality reduction techniques, including filter-, wrapper-, and embedded-based) and feature transformation techniques (e.g., principal component analysis (PCA) and Linear Discriminant Analysis (LDA)). Furthermore, we employed other machine learning and ensemble models such as Logistic Regression, K-Nearest Neighbour, Random Forest, Support Vector Classifier, Extra Trees Classifier, Gradient Boosting, Voting Ensemble, Stacking Ensemble, and a Hybrid Ensemble models, for the assessment of wine quality classification. Experimental results demonstrate that the correlation-based filter feature selection method achieved the best overall performance when combined with the proposed Hybrid Ensemble model, obtaining an accuracy of 89.08% and an F1-score of 88.93%. The findings indicate that dimensionality reduction combined with ensemble learning significantly improves prediction performance while reducing feature redundancy and computational complexity. The proposed approach provides an efficient and reliable framework for automated wine quality assessment and decision support in the wine industry.