Loan Default Prediction

• 

  Enabling logging

• 

  Application Insight

• 

  Project workflow

• 

  Comparing automl and hyperdrive model

• 

  Deployment state

• 

  Completed AutoML run

• 

  Model deployment

• 

  Details necessary for endpoint interaction

• 

  AutoML run metrics

• 

  Endpoint output

• 

  Best hyperdrive model

• 

  Endpoint response

• 

  Deleting service after use

• 

  Completed hyperdrive run

• 

  Sample data to test endoint

• 

  ONNX workflow

Inspiration

I own a degree in engineering but somehow developed interest in Machine learning and AI. Fortunately, this new skills of me propelled me into the financial space: I work as a data scientist in a financial institution. This, right here is my inspiration: finding myself in a field I never thought I'd be. It's a space I've grown to love and cherish. So, I thought of a way to add value by solving common problems that exist in the financial sector; one of which is loan default prediction.

What it does

The project identifies clients that are likely to default a loan credit.

How we built it

Project was built by training machine learning models on Azure ecosystem; leveraging tools such as automl, hyperparemeter tuning with hyperdrive, MLOps with pipelines, etc.

Challenges we ran into

Data privacy. Letting out clients information for the project was a challenge. A little simulation/editing of data was done to prevent valuable clients information from being released to the public.

Accomplishments that we're proud of

Project hit an accuracy of about 85% with AutoML. Model was deployed and consumed by exposing its RESTful endpoint API. Model was also converted to ONNX format, so it can be used on android and iOS devices

What we learned

There are so many resources on Azure to applicable to machine learning tasks.

What's next for Loan Default Prediction

Push it to the company's executive and see if it can be implemented in operations.

Built With

• automl
• azure
• hyperdrive
• mlops
• onnx
• python