deep_dive · 2021 — 2023

ML Loan Recommendation System

Developed machine learning recommender system to offer customized loans to over 500,000 borrowers, improving conversion rates.

k-means over borrower featuresschematick = 3·iter 00·inertia 0.000

Context

Offering every borrower the same loan product wastes both sides of the transaction. With a book of more than half a million borrowers, matching people to the product they would actually accept is a recommendation problem, not a marketing one.

Approach

Engineered custom borrower features from the underlying account and behavioral data
Applied K-Means clustering to group borrowers into meaningful segments
Designed similarity metrics to match each borrower against the loan products that fit their segment

Outcome

Deployed end to end for a base of 500,000+ borrowers
Improved conversion rates on loan offers

Stack

PythonMLscikit-learnRecommendation Systems