Identification of chemical markers to detect abnormal wine fermentation using support vector machines

Support Vector Machine (SVM) was explored as a tool for the early detection of abnormal fermentations, which are common in the wine industry. A database of about 18,000 data from 38 fermentations and 45 variables was used. Two cases were studied: (I) measurements of five groups (fermentation control variables, organic acids, amino acids, saturated and unsaturated fatty acids); and (II) four variables (density, YAN, brix and acidity). In addition, different kernels, training/testing configurations, and cut-off time were evaluated. Main results indicated that 80% of wine fermentations were well predicted using information of amino acids. In addition, density and YAN were the best individual chemical markers for prediction, with over 90% of accuracy at first 48 h of the process. Therefore, SVM can be used as a decision support tool for wine fermentation monitoring. Using data from the first 72 h, it is possible classify abnormal fermentations with high precision.