Bank Loan Fraud Pattern Analysis

A public sector bank identified suspicious patterns in 3,500 loans involving inflated property values and fake income documents. Total fraud amount is Rs. 500 crores. Students will analyze fraud indicators and develop risk scoring models.

# Aplly.xyz Case Study Submission

## Title
Bank Loan Fraud Pattern Analysis

## Type
Fraud Pattern Analysis

## Difficulty
Intermediate

## Estimated Time
60 minutes

## Overview
A public sector bank identified suspicious patterns in 3,500 loans involving inflated property values and fake income documents. Total fraud amount is Rs. 500 crores. Students will analyze fraud indicators and develop risk scoring models.

## Case Details

Function Focus: Fraud Pattern Analysis — manual reasoning, decomposition, and critical judgment (no spreadsheet or AI tool permitted in Phase 1)

Scenario:
You are the analyst at a fictional consultancy ("Praxis Advisors") tasked with answering: Bank Loan Fraud Pattern Analysis. You have a clean, synthetic dataset described below. You must produce a defensible answer using structured reasoning — no tool-assisted shortcut on the first pass.

Dataset Structure:
- application_id
- loan_type
- loan_amount_lakhs
- income_lakhs
- credit_score
- employment_type
- employer_verified
- discrepancy_count
- fraud_indicators
- status
- fraud_flagged

Tasks:
1. Profile the distribution of key numeric columns; identify outliers, tail values, or groups that look systematically different.
2. Flag candidate fraudulent records by rule-of-thumb thresholds you define yourself. Write the rules down before applying them.
3. Look for coordinated patterns (shared attributes, networks, repetition) that single-record checks would miss.
4. Estimate the financial exposure of the flagged set and state your confidence and the biggest assumption.
5. Only after the manual pass, compare your flags against an automated model/threshold and explain the gaps.

Expected Output:
A one-page fraud assessment memo: flagged record list with rules, network/pattern evidence, exposure estimate, confidence statement.

Evaluation Criteria:
Correctness of threshold rules, detection of coordinated patterns, accuracy of exposure estimate, honest confidence statement.

## Data Sources

| application_id | loan_type | loan_amount_lakhs | income_lakhs | credit_score | employment_type | employer_verified | discrepancy_count | fraud_indicators | status | fraud_flagged |
|---|---|---|---|---|---|---|---|---|---|---|
| L001 | Home | 50 | 12 | 620 | Salaried | True | 3 | Income mismatch, address fake | Rejected | True |
| L002 | Personal | 15 | 8 | 580 | Self-employed | False | 5 | ITR mismatch | Rejected | True |
| L003 | Business | 100 | 35 | 700 | Self-employed | True | 1 | None | Approved | False |
| L004 | Education | 10 | 5 | 650 | Student | True | 4 | College unverified | Rejected | True |
| L005 | Vehicle | 8 | 6 | 550 | Salaried | True | 2 | None | Approved | False |
| L006 | Home | 75 | 18 | 720 | Salaried | True | 0 | None | Approved | False |
| L007 | Personal | 25 | 10 | 510 | Self-employed | False | 6 | Bank statement tampered | Rejected | True |
| L008 | Business | 200 | 50 | 740 | Self-employed | True | 0 | None | Approved | False |
| L009 | Education | 8 | 4 | 600 | Student | False | 3 | Offer letter fake | Rejected | True |
| L010 | Vehicle | 12 | 9 | 680 | Salaried | True | 0 | None | Approved | False |


Full dataset: https://github.com/arora200/aplly_case_db/datasets/065-fraud-bank-loan (synthetic, 15 records)

## Solution Frameworks
Pattern & anomaly detection, network/graph analysis, risk scoring, layered defense review, distribution/tail analysis

## Solver Guidance & Tutorials
_Solver guidance added by the pipeline (tutorial links) — see `solver_guidance` field._

## What You'll Learn
- Spot outliers and coordinated patterns by hand
- Write and test your own detection rules
- Estimate financial exposure honestly

## Tags
Bank Fraud, Financial Crime, Pattern Analysis, Statistical Analysis

## Registration Links
Register as Solver / Register as Evaluator