Credit Risk Modeling with Real-Time Streaming Features: PD/LGD Pipeline Design

Jeevan Krishna Paruchuri

doi:10.36948/ijfmr.2023.v05i02.75348

Credit Risk Modeling with Real-Time Streaming Features: PD/LGD Pipeline Design

Author(s)	Jeevan Krishna Paruchuri
Country	United States
Abstract	Credit risk modeling for embedded lending in fintech neobanks requires real-time decision-making at transaction boundaries, where millisecond-level timing can determine whether a micro-credit request is approved or declined. The key metrics that drive credit decisions Probability of Default (PD) and Loss Given Default (LGD) depend on the quality and timeliness of features fed into them, and traditional monthly batch cycles create an unacceptable freshness gap. This paper presents a case study of a real-time streaming feature pipeline for PD/LGD models at a fintech neobank with a portfolio of more than 2.5 million microaccounts and a transactional volume of approximately 750,000 financial events per second on average, with 2 million events at peak during payment cycle waves. We describe the end-to-end architecture (Kafka ingestion from distributed payment APIs, Spark Structured Streaming feature computation across 1,100+ microservice event streams, Databricks Delta Lake persistence with audit logging, Feast feature store backed by Bigtable for ultra-low-latency retrieval, Seldon Core model serving on GKE with real-time prediction), the regulatory governance framework required for UDAP compliance and consumer protection, the feature drift detection and smoothing methods we use to manage the extreme volatility of microaccount spending patterns, and the operational realities of running this kind of pipeline at 99.8% measured availability against a 99.95% target. We compare batch and streaming approaches and argue for a hybrid: batch base features for slow-changing customer identity combined with streaming delta features for transaction-level behavior. We compare credit risk and fraud detection architectures in the fintech context, which share core streaming infrastructure but differ significantly in latency requirements and user experience impact. We close with a decision framework for fintech practitioners considering whether real-time PD/LGD features are worth the operational investment in a competitive lending market.
Field	Computer Applications
Published In	Volume 5, Issue 2, March-April 2023
Published On	2023-03-04
DOI	https://doi.org/10.36948/ijfmr.2023.v05i02.75348

View / Download PDF File

E-ISSN 2582-2160

doi

CrossRef DOI is assigned to each research paper published in our journal.

IJFMR DOI prefix is
10.36948/ijfmr

Downloads

Research Paper Format Copyright Permission Form and Undertaking Form Cover Page Vol 8 Isu 3 Cover Page Vol 8 Isu 2 Cover Page Vol 8 Isu 2

All research papers published on this website are licensed under Creative Commons Attribution-ShareAlike 4.0 International License, and all rights belong to their respective authors/researchers.

CC-BY-SA

About IJFMR Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics Peer Review & Plagiarism	Join as a Reviewer Editors & Reviewers Reviewer Referral Program Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us		+91-9687-828-838	editor@ijfmr.com

International Journal For Multidisciplinary Research

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Bi-monthly Scholarly International Journal

Credit Risk Modeling with Real-Time Streaming Features: PD/LGD Pipeline Design

Share this