K-Nearest Neighbors

Sampling

Splitting the data as training and test set, using createDataPartition( ) function from the caret package.

Preprocessing

KNN requires variables to be normalized or scaled. The caret package provides ways to preprocess data. Here we chose centering and scaling by settig preProcess = c("center", "scale")

Training and train control

Here we applied a 20-fold cross-validation as the trainControl, and trained the train data set with KNN methods. The output:

Plot the neighbors vs RMSE

Based on the KNN output and the plot, we choose K =23, with the smallest RMSE = 0.4573 and largest R_square = 0.3596.

Plot the KNN prediction vs Actual average rating