3.12 Summary
The objective of this chapter was to model one variable, wine quality, treating it as continuous, nominal, and ordinal. Depending on the type of the variable, we ended up using linear model, multinomial logistic regression, support vector machines, and ordinal regression. This exercise suggests that wine quality is not determined by the wine’s chemical composition alone.20
A common point of failure for MNL, SVM, and ordinal regression was that they incorrectly categorized a lot of quality 7 wines as quality 6 wines. Perhaps this suggests that there is not much difference between these two wines and the models are getting confused. Therefore, model accuracy can be improved if we combine these two groups together.
The original data set stripped out other relevant information such as price and brand of the wine. Perhaps including these variables will lead to a better model performance.↩