A key challenge to marketing campaigns is to target the customers that are likely to buy, and not waste time and money on those who are less likely to buy.

On Kaggle, there is a dataset from a Portugese bank’s marketing campaign. See Bank marketing campaigns dataset | Opening Deposit | Kaggle

We are here interested in the outcome of the phone call made to these customers: did they place a “term deposit” or not. This is found in the column “y”.

A small share of the calls made led to sales. It would be great if the bank could predict who will sign up so that the calls can focus on those customers and not having to call those who are predicted not to buy. (Although some might say that the job of the marketer is to convert the “no” to a “yes”..)

Can you create a predictive model that can help the bank save money by targeting the right people?

What Analytic technique did you use?

How accurate are your predictions?

In this use-case, what does false negatives and false-positives mean for the business? Are there any reasons one should be avoided more than the other? What influences this balancing act?

Hints

Use your favorite toolkit to try different analytic techniques. Weka makes it easy to try different methods!

Suitable approaches could be trees or logistic regression, or (this will take a bit longer to run) perhaps a k-Nearest Neighbour, a Multilayer Perceptron Neural Network or a support vector classifier (SMO in Weka)!