Detection of abnormal processes of wine fermentation by support vector machines

The early detection of problematic fermentations is one of the main problems that appear in winemaking processes, due to the significant impacts in wine quality and utility. This situation is specially important in Chile because is one of the top ten wine production countries. In last years, different methods coming from Multivariate Statistics and Computational Intelligence have been applied to solve this problem. In this work we detect normal and problematic (sluggish and stuck) wine fermentations applying the support vector machine method with three different kernels: linear, polynomial and radial basis function. For the training algorithm, we use the same database of 22 wine fermentation studied in [1, 2] that contains approximately 22,000 points, considering the main chemical variables measured in this kind of processes: total sugar, alcoholic degree and density. Our main result establishes that the SVM method with third degree polynomial and radial basis kernels predict correctly 88 and 85 % respectively. The fermentation behavior results have been obtained for a 80–20 % training/testing percentage configuration and a time cutoff of 48 h.