Success case
Prediction of the probability
of default for consumer credit
The challenge
A bank that has an existing model of predicting the consistency of payments when giving loans to costumers based off of certain easily obtained information.
Objectives
The goal is to provide a model capable of identifying customers with a higher probability of default more accurately. This way, the bank can grant loans to customers with greater confidence.
As loans will be repaid more regularly, the bank will be able to predict its future profits and expenses with greater accuracy.
How have we done it?
- XGBoost
- L2 Regulization
- Decision Trees
- Oversampling
- Predictive Model
Results provided
- Increase of the accuracy of predicting regular payment behavior by the applicant.
- Provision of a list of important features that determine payment security.
- Reduction of the risk of losses by selecting applicants with better performance.