Jigglypuff, is a Python package for creating a Risk Based Analysis Model. This package is based on customer segmentation, which involves categorizing the portfolio by industry, location, revenue, account size, and number of employees and many other variables to reveal where risk and opportunity live within the portfolio. Those patterns can then provide key measurable data points for more predictive credit risk management. Taking a portfolio approach to risk management gives credit professionals a better fix on the accounts, in order to develop strategies for better serving segments that present the best opportunities. Not only that, you can work to maximize performance in all customer segments, even seemingly risky segments.
Customer segmentation analysis can lead to several tangible improvements in credit risk management: stronger credit policies, and improved internal communication and cooperation across teams.
Jigglypuff package its comprehended by a Class called RiskDataframe The class is used to extend the properties of Dataframes to a particular type of Dataframes in the Risk Indutry.
It provides the end user with both general and specific cleaning functions, though they never reference a specific VARIABLE NAME.
It facilitates the End User to perform some Date Feature Engineering, Scaling, Encoding, etc. to avoid code repetition.
This package works setting as a target the column where are recorded the missing payments, this is our target and the prediction of the Logistic Regression model.
Jigglypuff can be pip installed, running the following command pip install Jigglypuff==0.2
SetAttributes function will update the type of the variable submitted for change. It will verify first that the key is present in the desired dataframe.
If present, it will try to change the type to the desired format. If not possible, it will continue to the next element.
kwargs : The key-argument pair of field-type relationship that wants to be updated.
Returns
None.
missing_not_at_random function will indicate if there is a relationship between missing data and its values.
Once the user has indicated in Jupyter notebook which features are pivot_value, target_value, down_payment, income_status, birth_date and dates today. The start function will proceed to clean the dataset in order to be suitable to prepare it for the logistic regression model.
_pivot_uniquefunction will be used to remove duplicates based on the pivot value assigned by the user._clean_targetfunction will convert the target values in a binary bucket, in order to reduce the complexity for the Logistic regression model._cleanfunction will clean the empty spaces with 'UNKNOWN'.agefunction will obtain the age from the birth date in the data frame, empty spaces will be filled with 'UNKNOWN'._downfunction will obtain the down payment from each, additionally the data frame is separated between individuals and corporate._incomefunction will simplify professions into unemployed or active in order to make professions an easier feature for segmentation._dayslapsedfunction will convert days into numerical values which are more useful for the model.
RiskDataframe class will automatically segment the categorical values in order to indicate if the split generates a better accuracy than the whole dataset. For categorical values it has been indicated a train of 10% and a test for the 90% of the data frame.
seg_data_cat is a list of columns which shall be specified by the user, in order to indicate the class which columns contains categorical values.
The method set_train_cat(target_value,seg_data_cat) will measure the effectiveness of splitting the categorical values of the data set.
Prior to execute the numerical variable analysis, the categorical values have to be converted to numerical values. In order to make this, encod function will run one hot encoder.
RiskDataframe class will automatically segment the numerical values in order to indicate if the split generates a better accuracy than the whole dataset. For numerical values it has been indicated a train of 10% and a test for the 90% of the data frame.
seg_data_num is a list of columns which shall be specified by the user, in order to indicate the class which columns contains numerical values.
The method set_train_num(target_value,seg_data_num) will measure the effectiveness of splitting the numerical values of the data set.
For both categorical and numerical values, the effectiveness is measured by GINI score.
!pip install Jigglypuff==0.2
from Jigglypuff.RiskDataframe import RiskDataframe as rdf
import pandas as pd
dataframe = pd.read_csv("AUTO_LOANS_DATA.csv", sep=";")
myrdf = rdf.RiskDataframe(dataframe)
myrdf.missing_not_at_random(input_vars=[])
The user has to set the pivot column (Loand contract number), the target value (Unpaid monthly installments), down payment, income status or profession of the lender, birth date and the rest of the dates on the dataset.
pivot_value = 'ACCOUNT_NUMBER'
target_value = 'BUCKET'
down_payment = 'PROGRAM_NAME'
income_status = 'PROFESSION'
birth_date = 'BIRTH_DATE'
dates_todays = ['REPORTING_DATE','LOAN_OPEN_DATE','EXPECTED_CLOSE_DATE','CUSTOMER_OPEN_DATE']
myrdf.start(pivot_value,birth_date,target_value, down_payment,income_status,dates_todays)
In order to set the data types a dictionary has to be created indicating which are the categorical variables.
argument_dict = {'PROGRAM_NAME':'category','SEX':'category',
'PROFESSION':'category','CAR_TYPE':'category'}
myrdf.SetAttributes(argument_dict)
myrdf.dtypes
The user has to indicate which are the categorical variables from the dataset.
seg_data_cat =['SEX','PROFESSION','CAR_TYPE','TYPE']
myrdf.set_train_cat(target_value,seg_data_cat)
The output of this method will indicate the GINI score for the full model and the segmented model, for the categorical variables, and will be indicated below whether the split model is better or worse than the full model.
(['The total accuracy using all variable and Logistic regression is: 0.8686868686868687',
'Using: SEX GINI Full Model Seg1: 30.303745053332285%',
'Using: SEX GINI Segmented Model Seg1: 30.303745053332285%',
'Using: SEX GINI Full Model Seg2: 19.219512195121947%',
'Using: SEX GINI Segmented Model Seg2:19.219512195121947%',
'Using: PROFESSION GINI Full Model Seg1: 26.49165350778253%',
'Using: PROFESSION GINI Segmented Model Seg1: 26.49165350778253%',
'Using: PROFESSION GINI Full Model Seg2: nan%',
'Using: PROFESSION GINI Segmented Model Seg2:nan%',
'Using: CAR_TYPE GINI Full Model Seg1: 47.844112769485925%',
'Using: CAR_TYPE GINI Segmented Model Seg1: 47.844112769485925%',
'Using: CAR_TYPE GINI Full Model Seg2: 22.12240785828228%',
'Using: CAR_TYPE GINI Segmented Model Seg2:22.12240785828228%',
'Using: TYPE GINI Full Model Seg1: 24.75394321766562%',
'Using: TYPE GINI Segmented Model Seg1: 24.75394321766562%',
'Using: TYPE GINI Full Model Seg2: 55.55555555555558%',
'Using: TYPE GINI Segmented Model Seg2:55.55555555555558%'],
['After analysis, we did not find a good split using: SEX',
'After analysis, we find a good split using: PROFESSION set at: ACTIVE',
'After analysis, we find a good split using: CAR_TYPE set at: HYUNDAI',
'After analysis, we find a good split using: TYPE set at: EMPLOYED'])
The user has to indicate which are the numerical variables from the dataset.
seg_data_num = ['ORIGINAL_BOOKED_AMOUNT','OUTSTANDING','BIRTH_DATE','DOWN_PAYMENT','REPORTING_DATE_DAYS_LAPSED','LOAN_OPEN_DATE_DAYS_LAPSED','EXPECTED_CLOSE_DATE_DAYS_LAPSED','CUSTOMER_OPEN_DATE_DAYS_LAPSED']
One Hot Encoder will be use to transform the categorical variables.
myrdf.encod(seg_data_cat)
myrdf.set_train_num(seg_data_cat,target_value,seg_data_num)
The output of this method will indicate the GINI score for the full model and the segmented model, for numerical variables, and will be indicated below whether the split model is better or worse than the full model.
(['The total accuracy using all variable and Logistic regression is: 0.8881829733163914',
'Using: ORIGINAL_BOOKED_AMOUNT GINI Full Model Seg1: 35.66609830232039%',
'Using: ORIGINAL_BOOKED_AMOUNT GINI Segmented Model Seg1: 35.66609830232039%',
'Using: ORIGINAL_BOOKED_AMOUNT GINI Full Model Seg2: 45.264030310206024%',
'Using: ORIGINAL_BOOKED_AMOUNT GINI Segmented Model Seg2: 45.264030310206024%',
'Using: OUTSTANDING GINI Full Model Seg1: 47.80138859667578%',
'Using: OUTSTANDING GINI Segmented Model Seg1: 47.80138859667578%',
'Using: OUTSTANDING GINI Full Model Seg2: 33.6887534907337%',
'Using: OUTSTANDING GINI Segmented Model Seg2: 33.6887534907337%',
'Using: BIRTH_DATE GINI Full Model Seg1: 39.5159865748101%',
'Using: BIRTH_DATE GINI Segmented Model Seg1: 39.5159865748101%',
'Using: BIRTH_DATE GINI Full Model Seg2: 34.40599069014372%',
'Using: BIRTH_DATE GINI Segmented Model Seg2: 34.40599069014372%',
'Using: DOWN_PAYMENT GINI Full Model Seg1: 34.9462365591398%',
'Using: DOWN_PAYMENT GINI Segmented Model Seg1: 34.9462365591398%',
'Using: DOWN_PAYMENT GINI Full Model Seg2: 36.48063200815494%',
'Using: DOWN_PAYMENT GINI Segmented Model Seg2: 36.48063200815494%',
'Using: REPORTING_DATE_DAYS_LAPSED GINI Full Model Seg1: 5.439161966156325%',
'Using: REPORTING_DATE_DAYS_LAPSED GINI Segmented Model Seg1: 5.439161966156325%',
'Using: REPORTING_DATE_DAYS_LAPSED GINI Full Model Seg2: 32.53278887081705%',
'Using: REPORTING_DATE_DAYS_LAPSED GINI Segmented Model Seg2: 32.53278887081705%',
'Using: LOAN_OPEN_DATE_DAYS_LAPSED GINI Full Model Seg1: 28.086180124223592%',
'Using: LOAN_OPEN_DATE_DAYS_LAPSED GINI Segmented Model Seg1: 28.086180124223592%',
'Using: LOAN_OPEN_DATE_DAYS_LAPSED GINI Full Model Seg2: 37.21047331319234%',
'Using: LOAN_OPEN_DATE_DAYS_LAPSED GINI Segmented Model Seg2: 37.21047331319234%',
'Using: EXPECTED_CLOSE_DATE_DAYS_LAPSED GINI Full Model Seg1: 1.2393767705382253%',
'Using: EXPECTED_CLOSE_DATE_DAYS_LAPSED GINI Segmented Model Seg1: 1.2393767705382253%',
'Using: EXPECTED_CLOSE_DATE_DAYS_LAPSED GINI Full Model Seg2: 26.236351958895334%',
'Using: EXPECTED_CLOSE_DATE_DAYS_LAPSED GINI Segmented Model Seg2: 26.236351958895334%',
'Using: CUSTOMER_OPEN_DATE_DAYS_LAPSED GINI Full Model Seg1: -2.7727046322430726%',
'Using: CUSTOMER_OPEN_DATE_DAYS_LAPSED GINI Segmented Model Seg1: -2.7727046322430726%',
'Using: CUSTOMER_OPEN_DATE_DAYS_LAPSED GINI Full Model Seg2: 38.69792852264968%',
'Using: CUSTOMER_OPEN_DATE_DAYS_LAPSED GINI Segmented Model Seg2: 38.69792852264968%'],
['After analysis, we find a good split using: ORIGINAL_BOOKED_AMOUNT set at: 92207.14078562526',
'After analysis, we find a good split using: OUTSTANDING set at: 33228.256089422415',
'After analysis, we find a good split using: BIRTH_DATE set at: 43.66400467967141',
'After analysis, we find a good split using: DOWN_PAYMENT set at: 0.40322744728997456',
'After analysis, we did not find a good split using: REPORTING_DATE_DAYS_LAPSED',
'After analysis, we find a good split using: LOAN_OPEN_DATE_DAYS_LAPSED set at: 2159.21330654391',
'After analysis, we did not find a good split using: EXPECTED_CLOSE_DATE_DAYS_LAPSED',
'After analysis, we did not find a good split using: CUSTOMER_OPEN_DATE_DAYS_LAPSED'])
By using Jigglypuff you agree to the following rules:
You must place the following URL in a footnote to help others find Jigglypuff: https://github.com/hana-am/GroupF_RBASegmentation#data-handling
You assume all risk for the use of Jigglypuff.
Alexander, J., Almashari, H., Galeano, A., Godinez, F., Hernández, T. (2021a).