For this question, suppose you are employed on the data analytics team for EZ-Mobile, a popular nation-wide cellphone provider. Your goal is to build a data mining system which helps customer support advise existing EZ-Mobile customers, and potential new customers, on which cellphone service plan is best for them (Plan A or Plan B). You have decided to predict each customers choice of plan using the K-nearest neighbor classifier on a dataset of the customers typical call behavior each day. A subset of the dataset is given in Table 1. The class attribute is the last column, Plan choice. The Customer ID attribute is not used for the predictions. Customer ID Local calls Long-distance calls Off-peak calls Plan choice (class) A 2 -1 1 Plan A B -1 2 0 Plan B C -1 0 -1 Plan A D 1 1 -3 Plan B E 1 1 -1 ? Table 1: EZ-Mobile Dataset

(a) Calculate the Euclidean distances between Customer E and each of the other customers, i.e. Customers A, B, C, and D.

(b) Based on the distances you calculated, which customer is the nearest neighbor for Customer E? Hence, what is the predicted class label for Customer E according to the nearest neighbor classifier?

(c) Based on the distances you calculated, which are the three nearest neighbors? Hence, what is the predicted class label for Customer E according to the K-nearest neighbor classifier, where K = 3?

(d) Explain what impact the choice of the number of neighbors K typically has on the behavior of the nearest neighbor classifier.

(e) Describe a sensible strategy for selecting the best value of K in the K-nearest neighbor classifier.