Predictive modeling for credit card offer acceptance

In this project we build a machine learning solution for analyzing customer behavior for credit card offer acceptance.

The Problem

This initiative is undertaken to delve into the trends in customer acceptance of credit card offers from financial institutions. Our aim is to enhance our comprehension of the factors influencing offer acceptances and to develop predictive insights into whether a customer is more inclined to accept or decline the offer.

The Data

More information on the problem and the data used for this project can be found here

The model

We used two models

• Logistic Regression
• Decision Tree The best model was Logistic Regression

How to Run The Project

Please note that these steps are to be carried out in a terminal window like command prompt or git bash

• Clone the repository to your computer using:
  

git clone https://github.com/Fozan-Talat/credit_card_ml.git

• Navigate to the directory where you cloned the repo and cd into the cloned repo by running:
  

cd credit_card_ml

You should now be in the project folder

• The libraries used in this project can be found in the requirements.txt file. It is advisable to run this project in a virtual environment to avoid issues with your system configuration. To create a virtual environment with venv
  

python -m venv credit-env

Activate credit-env with:

credit-env\Scripts\activate.bat

Now install the libraries from the requirements.txt file by running:

pip install -r requirements.txt

As we saw earlier, the model was already trained and saved in the train.py file. However, you may want to retrain the model for some reason. You do this by executing the file. In terminal, run:

python train.py

Now deploy the flask app locally by running the predict.py file. It is recommended to do this with a production server like waitress(windows OS) or gunicorn(UNIX).
You can still run it directly, with:

python predict.py

with waitress:

waitress-serve --listen=0.0.0.0:9090 predict:app

with gunicorn:

gunicorn --bind 0.0.0.0:9090 predict:app

Now the prediction service should be running locally on "http://0.0.0.0:9090"

Finally, you send a POST request to the service. You can do this by running the request.py file or use the file to understand the format in which the request will be sent, and then create your own request. This is how you run it:

python request.py

You can also decide to build and run the Dockerfile for the project. To build this locally:

docker build -t credit_card_project .

Then to run it:

docker run -it --rm -p 9090:9090 credit_card_project:latest