Wine (in this case red wine) has been graded for many years by experts who actually taste a sample of the wine, examine its color and aroma, and assign a grade (in our case, high quality or lower quality). Would it be possible, however, to use attributes of the wine that are machine measurable like fixed acidity and residual sugar to classify the wines.

Partition the data into two partitions (60% and 40% respectively), the training data and the validation data. Estimate a kNN algorithm for the training data and examine the resulting estimation.

Explain the overall misclassification rate for the validation partition and its calculation. What would have been the validation misclassification rate if you had used the naïve model? 

What does the validation confusion matrix tell the data scientist? Is the algorithm more likely to make one type of error rather than the other?

The lift chart is potentially the most important information provided by the algorithm. Examine either the cumulative gains chart, the lift chart proper, or the decile-wise lift chart. Do they each tell much the same story? How would you explain one of these lift charts to someone unfamiliar with predictive analytics?