https://nbviewer.org/github/IBM/AIX360/blob/master/examples/tutorials/HELOC.ipynb
loan applicant
For illustration purpose, we only represent a few features linked to the Protodash algorithm's prototypical example: https://nbviewer.org/github/IBM/AIX360/blob/master/examples/tutorials/HELOC.ipynb#prototypes.
Applicant 1
Prototypical Applicant 2
Prototypical Applicant 3
Prototypical Applicant 4
Prototypical Applicant 5
For illustration purpose, we only represent a few features linked to the Protodash algorithm's prototypical example: https://nbviewer.org/github/IBM/AIX360/blob/master/examples/tutorials/HELOC.ipynb#prototypes.
Applicant who served as good index - applicant 8
Applicant whose loan was denied
68.0
Average time loan application needs to be on file to flip credit decision for applicant
Average time, loan application is on file column
Boolean Rule Column Generator (BRCG)
Contrastive Explainer Method (CEM)
Pertinent positive output from CEM method for applicant 8
Explanation of first prototypical loan applicant
Explanation of third prototypical loan applicant
Explanation of fourth prototypical loan applicant
Explanation of fifth prototypical loan applicant
Contrastive explanation task to find contributors to loan applicant's credit risk
Asks, "What features minimally lead the applicant to be accepted and not rejected?"
Customer
Explanation for low risk performance
Data Scientist
Asks, "What are features that most impact low credit approval risk?"
Explanation of second prototypical loan applicant
External Risk Estimate Column
75.0
External Risk Estimate less than 75
85
External Risk Estimate of Prototypical Applicant 1
77
External Risk Estimate of Prototypical Applicant 3
83
External Risk Estimate of Prototypical Applicant 4
73
External Risk Estimate of Prototypical Applicant 5
81.0
External Risk Estimate Value necessary to turn decision
Task to find prototypical good applicants
280
Month since last trade open of good applicant index
82
Risk Estimate value for good applicant
Asks, "Who are good examples for low risk?"
HELOC Dataset
Loan Officer
379
Month since last trade open of prototypical applicant 2
338
Month since last trade open of prototypical applicant 3
789
Month since last trade open of prototypical applicant 4
230
Month since last trade open of prototypical applicant 5
Month since last trade open
Number of satisfactory trades value of good applicant
21
Number of Satisfactory Trades necessary to change applicant's loan decision
Number Satisfactory Trades Column
17.0
Number Satisfactory Trades less than 17
false
Output risk performance for applicant whose loan was denied
Output- Risk Perfomance Column
Percentage trades never delinquent
Percent trades never delinquent value for good applicant
Explanation that includes features that would flip a decision in favor of the applicant
Pertinent Negative Output from CEM
Asks, "What features if changed would flip the applicant's loan approval to be accepted?"
Explanation that includes features in support of applicant's low credit risk prediction
Protodash explainer - good examples
89
Risk estimate value of prototypical applicant 2
0.0
Risk performance - low