/Credit-Risk-Modelling

Primary LanguageJupyter Notebook

Credit-Risk-Modelling

Challenge faced by bank is wrt credit lending

Solution --> How to revise the current credit lending strategy of bank in a more scientific way.

About dataset

  • dataset is from CIBIL Bureau which consists of customers data
  • there are 61 independent feature and 1 dependent feature
  • dependent feature consists of 4 classes --> P1, P2, P3, P4
  • P1 is given to customer with very good credit score
  • P2 is given to customer which is good but not as good as P1 and so on.
  • Loan is also called TL (Trade line) in bank language.
  • Secured loans are something which is backed by a collateral while unsecured is not backed by anything.
  • This is why unsecured loans has more interest rates compared to secured loans

Hypothesis testing

Question - Are these two columns associated ?

  • Null Hypothesis (H0) --> this two are not associated

  • Alternate Hypothesis (H1) --> this two are associated

  • Alpha --> significance level, usually it is 5% or 0.05, This threshold tells us, main kitna galat ho sakta hoon.

  • Confidence level --> point estimate and margin of error (alpha), 1 - alpha

  • We need to find evidence to support the alternate hypothesis which we do using p-value.

  • If we don't have enough evidence then we can say that we don't have enough evidence to support alternate hypothesis, and if we have enough evidence then we reject the null hypothesis.

  • p-value is calculated using tests, tests like t-test, chi-square test, Anova.

  • if p-value <= alpha --> Reject H0 and if p-value >= alpha --> fail to reject the H0

TESTS

  • Chisquare - categorical v categorical

    Ex : marital status vs loan approved or not

  • t-test - categorical v numerical (2 categorical columns) --> only 2 value_counts

    Ex : Age vs loan approved or not

  • ANOVA - categorical v numerical (>=3 categorical columns) --> more than or equal to 3 value_counts

VIF - Variance inflation factor

  • Used to identify multicollinearity
  • Takes r-squared value and eliminate if crosses the threshold
  • 1/ 1 - r-squared
  • VIF = 1 --> no multicollinearity
  • VIF = 1 - 5 --> low multicollinearity
  • VIF = 5 - 10 --> moderate multicollinearity
  • VIF > 10 --> High multicollinearity

Explaining to Business --> Interpretation of the outcomes

  • It depends on Risk appetite
  • If Risk appetite is low --> they are not willing to take much risk --> we can say target only P1 customers
  • If Risk appetite is high --> they are willing to take risk --> we can say target P1, P2 and maybe P3