Question 2

Suppose the bank is interested in classifying its customer’s value. As the data scientist of the bank, you suggest it to use a machine learning model to automate this task. Therefore, you collected a data set of 20 customers, as shown in the below table. If a customer’s value is higher than a threshold, the label of this customer is ‘Good’; otherwise, the label is ‘Bad’.

(a) (Python code) Suppose you would like to use the first 16 data points in the table as the training dataset, and the last 4 data points as the validation dataset. Implement the KNN classifier with K=3 and report the accuracy of your model on the validation dataset.

(b) (Python code) Propose at least one way to improve the performance of the KNN classifier developed in Q(a). You should report the improved accuracy on the validation dataset and explain your method. [Remark: the parameter of K is fixed as 3]

(c) Apart from the KNN method, you can also use a neural network for the above classification task. In the context of this question, which model do you think is more appropriate to use in this problem and why?