Using data mining techniques to predict industrial wine problem fermentations

摘要: Abstract Winemakers currently lack the tools to identify early signs of undesirable fermentation behavior and so are unable take possible mitigating actions. Data collected from tracking 24 industrial fermentations Cabernet sauvignon were used in this study explore how useful is data mining detect anomalous behaviors advance. A database held periodic measurements 29 components that included sugar, alcohols, organic acids amino acids. Owing scale problem, we a two-stage classification procedure. First PCA was reduce system dimensionality while preserving metabolite interaction information. Cluster analysis (K-Means) then performed on lower-dimensioned group into clusters similar behavior. Numerous classifications explored depending used. Initially just first three days assessed, entire set Information days’ provides important clues about final classification. We also found strong association between problematic specific patterns by tools. In short, contain sufficient information establish likelihood finishing normally. Results most encouraging. many more different varieties needs be collected, however, develop reliable broadly applicable diagnostic tool.