Behavioral Episode Learning for Credit Decisioning: Sequential Pattern Extraction and Life Event Detection from Transaction Histories

Read full text Summarise with VIRAT
Authors: Anil Kumar Sharma
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Traditional credit scoring systems such as CIBIL and Experian assign creditworthiness through static, point-in-time aggregates computed over historical repayment records. This methodology excludes over 400 million Indian adults who lack sufficient formal credit history — the so-called thin-file population — and fails to capture the dynamic behavioral trajectories that predict future repayment capacity more reliably than any static snapshot. We present Behavioral Episode Learning (BEL), a framework implemented within the ankrBFC (Behavioral Finance Cloud) platform that extracts sequential behavioral episodes from transaction histories to construct a time-aware credit assessment model. The system employs sliding-window episode extraction across five behavioral dimensions — salary regularity, bill payment timing, discretionary spending volatility, emergency withdrawal patterns, and investment consistency — and feeds episode sequences into an LSTM-based scoring architecture. Complementing credit assessment, we introduce a privacy-preserving life event detection engine that identifies marriage, job change, health crises, and home purchase events from Merchant Category Code (MCC) signatures without accessing transaction descriptions or any personally identifiable information. Decision explainability is provided through SHAP (SHapley Additive exPlanations) decomposition, satisfying RBI Fair Practices Code requirements. A continuous learning framework retrains monthly on loan outcome data and includes concept drift detection to handle regime shifts. Together, these components address credit exclusion at population scale while remaining compliant with the Digital Personal Data Protection (DPDP) Act 2023.

Fields of Study & Subjects

Financial Distress and Bankruptcy Prediction Privacy-Preserving Technologies in Data Microfinance and Financial Inclusion

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
25 March 2026
Language
EN
Document type
Preprint