Comparison of Machine Learning Algorithms in the Prediction of Pisco Varieties Using Near-Infrared Spectroscopy (NIRS)

The objective of the research was to perform a comparison of machine learning algorithms to identify the varieties of Peruvian pisco using near-infrared spectroscopy (NIR). Twelve samples of Peruvian pisco were used, including different types of green must from which spectral data were obtained. The data were then processed, and the dataset was obtained. Classification techniques were applied using the K-Nearest Neighbors (KNN) and Naive Bayes models. The results reveal that Naive Bayes, with 55.9%, presents a slight superiority over KNN, with 50.8%, in terms of accuracy and ability to identify the Peruvian pisco variety. The Recall, F1 Score, and Precision metrics support this superiority by demonstrating Naive Bayes’ ability to avoid false positives and adequately capture relevant cases. This approach integrates NIR spectroscopy and machine learning, demonstrating its effectiveness in the authentication of Peruvian pisco varieties. These findings underscore the importance of this technology in maintaining the quality and uniqueness of Peruvian pisco in international markets, consolidating its position as a vital tool in preserving the product's identity.