Benchmarking Machine Learning Classifiers for Wine Cultivar Identification from Chemical Analysis: A Reproducible Evaluation on the UCI Wine Dataset

Abstract Identifying the grape cultivar behind a wine from its chemical composition is a classic multi-class pattern-recognition task relevant to food authenticity and quality control. This study benchmarks logistic regression, a radial-basis-function support vector machine (SVM), k-nearest neighbors, random forest, and gradient boosting on the UCI Wine dataset distributed with scikit-learn, which contains 178 samples, 13 chemical measurements, and three cultivar classes. A stratified 80/20 train–test split was fixed with random seed 42, and five-fold stratified cross-validation was run on the training set. Logistic regression, RBF SVM, and k-nearest neighbors each reached 97.22% held-out accuracy, random forest reached 100%, and gradient boosting reached 94.44%. Because the test set contains only 36 samples, one prediction changes accuracy by 2.8 percentage points, and the cross-validation results rank the models differently. These results show strong class separability on a small, well-known benchmark; they do not establish performance on new vintages, regions, or laboratories.