This experiment shows how Credit Risk prediction can be done on corporate suppliers and customers using Machine Learning.
Scenario
--------
The company in this experiment need to score their corporate suppliers and customers by analyzing their balance sheet. We are given pre-processed financial data of companies that stayed healthy or went bankrupt over the course of 1 year.
The task is to build a machine learning model that will learn from the given historical data, and be able to predict from financial data which company would be unable to pay their debts one year from now.
Data Sets
---------
![Corporate Credit Dataset][1]
Each row in this dataset represents one company, and the columns X1, X2, etc... are all figures from the company's financial statements that have already been cleaned and statistically pre-processed for you.
Content
-------
This demo shows how to
1. Build a classification model to predict bankruptcy vs no bankruptcy
2. Deploy the model to Excel
Tutorial (from [Demand Forecasting tutorial][2])
--------
Please see the tutorial on Demand Forecasting for a step by step tutorial on how to build and deploy models in Azure ML Studio.
Available here on this [youtube playlist][3]
Outcome
-------
We can see that by using a two-class Decision Forest Classification, we were able to automatically pick out the elements in a financial statement that signified that a company was going to go bankrupt 12 months in future with 99% recall.
Learning point: Note that the number of total samples available was low, and that upsampling of the unbalanced class was conducted. This could have led to a better an expected result, and we might not see the same results with real world data that has more noise.
[1]: https://raw.githubusercontent.com/johnangrs/aisgbricks/master/data_viz_corp_credit.png
[2]: https://gallery.azure.ai/Experiment/Demand-Forecast-of-Time-Series-Data
[3]: http://youtube.com/playlist?list=PL_hc21tLhYNhCPablkp0sHzwC4P9FH6YQ