Product Quality Analysis Using Support Vector Machines

This paper presents an exploratory study of the effectiveness of support vector machines in the prediction of a product quality based on its characteristics. The study answers the following three questions: how does the choice of kernel and model parameters affect the predictive abilities of support vector machines; can an alternative subset of variables be unearthed that can be used in order to increase the predictive abilities of the data mining model; how will the removal of potential outliers affect the predictive abilities of the data mining model. We used a dataset of red and white wine samples presented by their physiochemical characteristics. Findings show that a correct selection of kernel and appropriate variable selection technique may have a significant impact on the prediction ability of the data mining model. Certain model settings can even make it to outperform the best technique reported thus far in the application area.