Continuing the previous problem, the same data have been split into two sets in the file P17_10.xlsx. The first 9500 observations are in the Training Data sheet, and the last 500 observations are in the Prediction Data sheet. In this latter sheet, the values in the Catalog Purchase and Online Purchase columns have been deleted. The purpose of this problem is to use Neural-Tools to train a neural net with the data on the first sheet and then use this neural net to predict values on the second sheet. Proceed as follows.

a. Designate a Neural-Tools data set for each sheet. The Customer and Catalog Purchase columns should be marked Unused, and the Online Purchase column should be marked Category Dependent. (The Catalog Purchase column is ignored in this problem.)

b. Use the Neural-Tools Train option to train a neural net on the first data set, using the PNN algorithm. You can accept the option to set aside 20% of the 9500 observations for testing. Then interpret the outputs. In particular, can you tell how the neural net making its predictions?

c. Use the Neural-Tools Predict option to predict Online Purchase for the observations in the Prediction Data sheet. What can you say about the resulting predictions? If you were forced to choose some people as most likely to make an online purchase, which people would you choose?