The use of data mining to classify Carménère and Merlot wines from Chile. Issue 2 (11th December 2018)

• Record Type: Journal Article
• Authors: da Costa, Nattane Luíza; Llobodanin, Laura Andrea García; Castro, Inar Alves; Barbosa, Rommel
• Is Part Of: Expert systems. Volume 36:Issue 2(2019)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2018-12-11
• Subjects: Carménère -- data mining -- feature selection -- Merlot -- support vector machines -- wine classification; Expert systems (Computer science); 006.33
• DOI: 10.1111/exsy.12361
• Languages: English
• ISSNs: 0266-4720

Abstract

Abstract: The identity of the Carménère variety was lost in Chile and regarded as a Merlot grape. This grape disappeared in Bordeaux, France, the country of origin of the grape, because of the phylloxera plague. In the present paper, a study on the classification of Chilean Carménère and Merlot wine samples based on chemical parameters was carried out. A total of 64 samples were analysed, and 20 chemical parameters were determined. Forty‐five samples were labelled as "Carménère" and 19 samples as "Merlot" according to the wine label. The samples were preprocessed with sampling algorithms to double the number of Merlot samples to 38 and reduce the Carménère samples to 38. The dataset was analysed with the data mining techniques support vector machines and correlation‐based feature selection. The capability of classifying the samples with all 20 chemical parameters was 86.8% accurate, and when using only the variables a*, total anthocyanins, cyan‐3‐glu, malv‐3‐acetylglu, peon‐3‐acetylglu, and vitisin A, which were selected through correlation‐based feature selection, accuracy increased to 93.4%. Therefore, wine anthocyanins and the parameter a* of wine colour proved useful to discriminate the wines.