Classification project

The objective of this project was to is to build a model that will provide insight into why some bank customers accept credit card offers, as well as answer additional questions by top management related to the matter.

In ordert to do this, I used agile project planning in Github and made use of SQL, Python and Tableau.

About the project

For the project we have data from a marketing study with 18.000 current bank customers. We want to understand the demographics and other characteristics of its customers that accept a credit card offer and that do not accept a credit card. .

The goal of the binary classification project is to analyse the characteristics of the bank customers and to train a model to predict if a bank customer accept or reject a credit card offer.

Dataset

For the project we have data from a marketing study with 18.000 current bank customers with the following columns:

Column	Information
customer_number	A sequential number assigned to the customers.
offer_accepted	Did the customer accept (Yes) or reject (No) the offer.
reward	The type of reward program offered for the card.
mailer_type	Letter or postcard.
income_level	Low, Medium or High.
bank_accounts_open	How many non-credit-card accounts are held by the customer.
overdraft_protection	Does the customer have overdraft protection on their checking account(s) (Yes or No).
credit_rating	Low, Medium or High.
credit_cards_held	The number of credit cards held at the bank.
homes_owned	The number of homes owned by the customer.
household_size	Number of individuals in the family.
own_your_home	Does the customer own their home? (Yes or No).
average_balance	Average account balance (across all accounts over time). Q1, Q2, Q3 and Q4
q1-q4 balance	Average balance for each quarter in the last year

Workflow

1. Exploring the data in SQL

   • Files: Solutions_SQL- Classification.ipynb / Questions_SQL - Classification.md /
   • Write queries to extract demographics and other characteristics of bank customers

2. Logistic regression in Python

   • Files: Solutions_Python - Classification.ipynb / helper.py / Questions_Python - Classification.md
   • EDA
   • Data processing, feature engineering
   • Model evaluation
   • Overview - model results

3. Analyse and visualize the data in Tableau

   • Files: Solutions_Tableau- Classification.ipynb / Tableau.twb / Questions_Tableau- Classification.md
   • Extract demographics and other characteristics of bank customers and visualize them

Conclusions

Note: For futher details, please refer to the related files

1. Exploring the data in SQL

• Properties: Credit rating medium or high, Credit cards held 2 or less, Owns their own home, Household size 3 or more
• There are only 167 customers with the following properties who accepted the offer. So I would give the hint, that this customers are under 1% of all customers and that is worth addressing this customer group.

2. Logistic regression in Python

• Logistic Regression with changed class weights fits best for this dataset. Highest Yes recall: 0.69.
• Next step to evaluate this model: Cut the variables which do not improve the prediction and improve the weight/balance
• For the next marketing study I would recommend to change the questions which have no relationship to the target variable (Like shown in point 5.2.), target people who live in a household of the size 3 or 4, and to build bins (For example house hold size 5-9)

3. Analyse and vizualize the data in Tableau

• There is a huge jump in average balance from Q1 to Q2 for households with size 8.
• The jump is caused by the fact that only one out of 18.000 customers has a household size of 8 and this particular person has an unusually high balance.

Libaries

• helper_classification
• IPython
• pandas as pd
• scipy
• seaborn as sns
• matplotlib
• numpy as np
• imblearn
• sklearn
• warnings