k and Feature Selection

These are feature selection charts paneled by k, and show how the selected model was chosen. Points in each chart display the error rate on the y-axis for the model with the feature set at each step of the feature selection process on the x-axis. Recall that the feature selection method was to choose the 5 "best" features. Thus, while the model with k=4 at step 2; that is, with 2 features added, has the lowest overall error, the model selection process chooses the model with the lowest error at step 5. This is the model with k=3.

You may want to consider creating a model with k=4 and the two features in the model at step 2 in order to make a closer comparison between it and the selected model. You can obtain the name of the variable added at each step by hovering over the point in the plot.