LTFS-Data-Science-FinHack.

The objective of this project is to predict the probability of borrower defaulting on a vehicle loan in the first EMI (Equated Monthly Installments) on the due date.

Overview:

End-user: Financial institution

Objective: The objective of this project is to predict the probability of loanee or borrower defaulting on a vehicle loan in the first EMI (Equated Monthly Instalments) on the due date.

Dataset: The Vehicle Loan Default Prediction dataset includes following features:

Dependent feature: loan_default

Independent features: disbursed_amount, asset_cost, ltv, PERFORM_CNS.SCORE etc.
The dataset contains 345550 rows and 41 features.

Data Cleaning:

• Combining training and testing data.
• Checking missing values in data.
• Imputing Employment.Type missing values with ‘Self employed’ values.
• Checking the outliers using boxplot.
• Removing the outliers of ‘disbursed_amount’ & ‘asset_cost’ features.

Feature Enginnering & Feature Selection:

• Calculating Age column using ‘Date.of.Birth’.
• Calculating ‘AVERAGE.ACCT.AGE’ and ‘CREDIT.HISTORY.LENGTH’ in months.
• Creating bins of PERFORM_CNS.SCORE and LTV.
• Replacing Values of PERFORM_CNS.SCORE.DESCRIPTION.
• Generating New Features like ‘ACTIVE.ACCTS’,’CURRENT.BALANCE’ etc.
• Dropping irrelevant columns like ‘DisbursalDate’, ‘Current_pincode_ID’ ,’NO.OF_INQUIRIES’ etc.

Model Creation:

CatBoost Classifier: ROC AUC Score: 0.6442

AUC ROC:

Thank You!

@ShehzadaAlam on GitHub">Follow @ShehzadaAlam