Credit card defaults -HCL

Inspiration

Credit card default is the serious problem in banks that affects their business growth. Is it possible to know the credit card defaults in early stage? obviously Yes

What it does

It is identify credit card defaults in early stage & then business can take appropriate actions to avoid defaults. Using data science,it helps to identify factors & associate relationships by using historical data.

How I built it

First performed descriptive analysis to identify relationships between variables & to know how these variables affected for target variable. After that performed various machine learning classification models & statistical techniques(sampling,PCA) to get better accurate model for given data set. Used azure machine learning studio notebook version.

Challenges I ran into

This is the unbalanced data set because defaulted was 22.12% & not defaulted was 77.88%. Overall model performed sufficient accuracy but category wise recall values were insufficient in first stage because of the unbalanced issue .

Accomplishments that I'm proud of

To get scalability & accurate model, performed various data engineering techniques. To solve unbalanced issue,performed various sampling techniques & final best technique is SMOTE. Finally identified important features using advanced analysis.

What I learned

I Learned various statistical sampling techniques & azure cloud techniques.

What's next for Credit card defaults -HCL

Next step is improving this model,by researching & applying other suitable variables. Applying various deep learning techniques such as deep neural networks & auto encoders.

Built With

• azure
• machine-learning
• python
• statistics