Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/fedesgh/building_credit_risk_classifier_using_bagging_kneighbors

Problem statment about modeling target vector and attempt to improve metrics
https://github.com/fedesgh/building_credit_risk_classifier_using_bagging_kneighbors

feature-selection imblearn information-value sklearn

Last synced: about 1 month ago
JSON representation

Problem statment about modeling target vector and attempt to improve metrics

Awesome Lists containing this project

README 

        
## Motivation



The motivation for this repository are the difficulties that the dataset present when we define the Target and Features. One of the problems involve **several data leakages**. 



There are several attempts in kaggle with **low metrics** particularly when we restrict the training set to features with information before the loan was granted and we want try to improve it:



https://www.kaggle.com/datasets/devanshi23/loan-data-2007-2014/data 



We use various data preprocces techniques like **SelectKbest with information value**, **Binning** , **Up-sampling with Imlearn**, **One Hot Encoder** and **Imputers**



## Problems at defining the target



**loan_status** (our target) has the followings values: 

    

  1. Current                                                
    2. 

  2. Fully Paid                                             
    3. 

  3. Charged Off                                             
    4. 

  4. Late (31-120 days)                                       
    5. 

  5. In Grace Period                                          
    6. 

  6. Does not meet the credit policy. Status:Fully Paid       
    7. 

  7. Late (16-30 days)                                        
    8. 

  8. Default                                                   
    9. 

  9. Does not meet the credit policy. Status:Charged Off       
    10. 




**The main point we must consider is that the values belong to differents moments in the loan life span.** 



Those that belong to an end of the Loan:

    

  1. Fully Paid                                             
    2. 

  2. Charged Off                                             
    3. 

  3. Does not meet the credit policy. Status:Fully Paid       
    4. 

  4. Default                                                   
    5. 

  5. Does not meet the credit policy. Status:Charged Off       
    6. 




Middle term of a loan:



    

  1. Current                                                
    2. 

  2. Late (31-120 days)                                       
    3. 

  3. Late (16-30 days)                                        
    4. 




while In Grace Period belongs to the beginning. 



On top of this we should consider:



**All the loans regardless its end, were previously in time "In Period Grace"**



**All the loans regardless its end, were previously in time Current and/or Late**



## Our target 



"Good loans" **(1)**: 

    

  1. Fully Paid                                             
    2. 




"Bad loans" **(0)**: 



    

  1. Charged Off                                             
    2. 

  2. Does not meet the credit policy. Status:Fully Paid       
    3. 

  3. Default                                                   
    4. 

  4. Does not meet the credit policy. Status:Charged Off       
    5. 




We just consider ends of loans categorys in the target, and we should consider only features in X_train set that belong **before**

the loan was granted.



## Result metrics.



![result.jpg](result.jpg)