Transcript slides

Predicting Mortgage Pre-payment
Borrower pays off the loan before the contracted
term loan length.
• Lender loses future part of the income stream
associated with the loan.
Inspiration & Previous work
Oded Netzer’s talk on text-mining techniques for
business applications – loan default prediction.
Breakdown of the Problem
To pin-point among borrower profiles, who all likely
to prepay the mortgage.
UCI Machine Learning Repository :
Credit Approval Dataset – 690 observations; 15
German Credit Dataset – 1000 observations, 20
Give Me Some Credit - Kaggle credit-scoring
competition - very large.
Prospective Solutions
Techniques for categorization of data
Decision Tree - recursively separates observations in
branches to construct a tree to improve prediction
accuracy, with use of measurements like information
gain ratio, Gini index etc.
 Naïve Bayes Classifier - calculates a set of
probabilities by counting the frequency and
combinations of values in the dataset.
 Logistic Regression - measures the relationship
between the one dependent binary variable and one
or more independent variables. by estimating
probabilities using a logistic function.Appropriate
when the dependent variable is binary.
Evaluation of Results
Measure of success
How clear the proposed categorization scheme is
How applicable the scheme is in enabling lenders in
managing prepayment risk by providing a useful
structure for early mitigation targeting.