Telemarketing-Prediction-for-Banking

UPDATE:

Please use the following link to open the notebook.

Go to notebook

There is some issue with the notebook viewer on Github

This is an independent Jupyter notebook where we want to predict whether a customer will buy a term loan or not

About the dataset

It is a dataset, describing the results of marketing campaigns run by a Portuguese bank. Conducted campaigns were based mostly on direct phone calls, offering clients to place a term deposit. If after all marketing efforts, client had agreed to place the deposit - target variable is marked yes, otherwise no

Tasks

1. Reducing the costs of the marketing teams incurred during campaigns
2. Improving the success rate of the converting the prospect to customer

Gains Chart and Lorenz Curve

1. N is the size of each of my deciles (10% of the original population).
2. Event Rate and Non Event Rate will add upto 100%.
3. KS is the difference between Cum ECR and Cum NECR. We want to check where it maximizes (yellow shaded cell).
4. In the No Model Scenario, we are taking the standard assumption that each decile is only contributing to 10% event capture. This will be useful for the baseline comparision
5. Hence, we can target the top 2 deciles to capture (Total Event Capture Rate => 46 +18 => 64%).
6. The maximum discrimantion is achieved in the top 2 deciles but we can also move upto the 3rd decile.

If we were to target the top 3 deciles {10, 9, 8} then we would captured 71% compared to 30% in a no model scenario

We can also design audience profiles and perform prioritze targeting to improve overall perfomance when compared to random selection of audience

TODO

1. Audience Profile creation