Use of Decision Tree Analysis for Determination of Critical Enological and Viticultural Processing Parameters in Historical Databases

Abstract

A method for optimizing wine quality or characteristics based on historical enological and viticultural data is being developed. The first step in this method is a database analysis and classification step where the most critical inputs to the process are determined. To identify critical processing inputs in large databases, we explored the use of decision tree analysis in conjunction with information theory for data classification. The success of this method lies in its ability to treat "categorical" variables (for example, vineyard, clone), which are typical of archived viticultural and enological data, as well as "continuous" variables (for example, temperature, sugar concentration). We have demonstrated that decision trees identify important variables using an enological database for which statistical analysis was previously completed. Results from both decision tree analysis and statistical analysis agree well. Both find the same subset of winemaking inputs in this database to be significant in classifying outputs such as tartaric acid concentration, total phenolic concentration, glycerol concentration, and malolactic fermentation duration. Decision tree analysis has also been tested successfully on an actual viticultural database from a winery to predict the effect of viticultural practices on wine quality, reducing a database of twenty inputs to six inputs that had a significant effect on a measure of wine quality. Partial Least Squares analysis has been explored as an alternative classification approach for databases containing only continuous variables, yielding comparable results to decision tree analysis. The final method developed will provide wineries with a straightforward means to identify their most important enological and viticultural practices and ensure production of a wine with the desired qualities.