Classification Models for Credit Underwriting

Project was aimed at identifying credit worthiness of a customer. I used multiple data sources, Bureau history, Bank Statement, SMS history, Apps installed on phone and Payment history on existing loans on self book. I cleaned and wrangled the data, created over 2500 features depicting different facets one's financial literacy, stability, spending and lifestyle habits. I used advanced dimensionality reduction techniques and classification algorithms to train an ensemble model with >78% AUC. I deployed the model using AWS Sagemaker improving portfolio risk spread by over 1200%.