Partner

Blog

HelpDesk

Banking-Machine Learning and AI
Banking

Banking

Overview – Analytics and Modeling Approach

Overview – Analytics and Modeling Approach

Business Understanding and objective definition

Business understanding

  • Research on banking industry with a specific focus on credit card business
  • Data understanding and requirement gathering
  • Understand the currently deployed predictive models deployed (if any)

Objectives

  • Predict if a credit card customer will churn ?
  • Assess the reason why the customer will churn?
  • What intervention can be provided to stop the customer from churning?
  • Conduct similar analysis for customer delinquencies

Data exploration and treatment

Data exploration and treatment

Data parameterization

Rank ordering predictor variables by significance metrics

Information value

Captures the relevance of a predictor variables using entropy concepts based in information theory

Variable Importance

Estimates whether randomly permuting values in a predictor variable decreases prediction accuracy

KS Statistic

Calculates maximum distance between the cumulative good and bad curves when a predictor variable is sorted and binned

Multi-colinearity

Only one predictor variable among a highly correlated group of predictor variables is allowed in the model, thus eliminating noise and redundancy.

Shortlisted variables for Model building

Shortlisted variables for Model building

Evaluate different models to select the best fit

Evaluate multiple traditionally used and high performance new age models

Evaluate multiple traditionally used and high performance new age models

Selection criteria

  • Complexity in model building
  • Efficiency of model
  • Avoid overfitting
Implement the best fit model

Implement the best fit model

Customer ID Age Location changed …. Churn Probability
12345679082 51 Yes 0.02
76832078294 33 Yes 0.61
13462394930 21 Yes 0.52
23746239242 24 No 0.34
  • Probability of churn at a customer level
  • Why will a customer churn?
  • When will a customer churn?
  • Value at risk for a churning customer?
  • Common characteristics among high risk customers
  • Intervention to prevent churn

Model enhancement can then be done using:

  • Social media data integration
  • Perform Network analysis