Homework answers / question archive / 1) In this folder, there is a dataset “credit_risk_test” and the python file “biasInMachineLearning

1) In this folder, there is a dataset “credit_risk_test” and the python file “biasInMachineLearning

Computer Science

Share With

---

1) In this folder, there is a dataset “credit_risk_test” and the python file “biasInMachineLearning.ipynb”. In this python file below steps followed are-

• Prepare dataset
• Create test and train splits
• Calculate actual disparate impact -testing values
• Train a model
• Evaluate performance
• Calculate actual disparate impact -predicted values
• Apply Disparate Impact Remover- AIF360 tool

 

1. Details of Dataset:

• Loan ID
• Gender (Male or Female)
• Race (Caucasian, African-American, Native American, Asian, Hispanic, Other)
• Married (Yes or No)
• Dependents (0, 1, 2, or 3+)
• Education (indicating whether or not the primary applicant has graduated from high school)
• Self Employed (Yes or No)
• Applicant Income
• Co-Applicant Income
• Loan Amount
• Loan Applicant Term
• Credit History (0 or 1, with 0 indicating good credit history)
• Property Area (Rural, Semiurban, or Urban)
• Loan Status (Y or N)

 

 

Right now, the bias and fairness is visualized as below in IBM AIF360 tool:

 

 

 

 

For detailed information about IBM AIF360 tool please refer:

https://aif360.mybluemix.net/

 You can access the API, code, demo, tutorials etc

 

 

 

 

 

TASK: I have to create visualizations that looks like below image. In the sense, it should show the different fairness metrics (From IBM tool i.e., Disparate Impact, Average Odds Difference, Equal Opportunity difference, Statistical parity difference) and a similar visualization. This visualization is interactive in nature. When the person clicks on the button the visualization changes accordingly.

For this visualization, the attributes can be converted into binary values.

 

Any libraries like D3.js, Bokeh, Plotly along with Python can be used.

 

Please refer the documentation of below visualization here-

https://research.google.com/bigpicture/attacking-discrimination-in-ml/

 

 

 

 

 

NOTE: The color and design can change accordingly. Or even a new kind of visualization similar to this is okay. It is not an issue. But I am mainly looking for similar visualization where one can identify bias for each fairness metric and is interactive.