The-House-Of-Kgosi/The-Paper-Claim

Insurance fraud detection system for the South African market. Five engines analyze policy inception timing, claimant networks, repair inflation, and temporal anomalies to flag suspicious claims. Produces ranked SIU referral queue with 0.87 precision. # The Paper Claim **Insurance fraud detection — because the most expensive accidents are the ones that never happened.** > Built by KM Mokoena · South Africa --- ## The Problem The Association for Savings and Investment South Africa (ASISA) estimates that fraudulent and dishonest claims cost the SA insurance industry over R1.9 billion annually. The real figure is higher — most fraud is never detected, absorbed into loss ratios and passed to honest policyholders through premium increases. The patterns are consistent. Claims submitted days after policy inception. Multiple claimants from the same incident sharing address history. Repair shops submitting inflated round-number quotes. Policyholders with three claims in two years. Staged accidents where the incident date precedes the report date. No single signal is proof. But statistically, the accumulation of signals separates legitimate claims from fraud. **The Paper Claim** ingests policy, claim, claimant, and repair data, runs five validated detection modules, and produces a per-claim fraud probability score with a ranked SIU referral queue. --- ## Detection Modules ### 1. Inception Proximity Engine (`src/inception_engine.py`) Short policy tenure before a high-value claim. | Signal | Method | |--------|---------| | Claim within N days of inception | Days-to-claim threshold | | Premium-to-claim ratio anomaly | Premiums paid vs claim value | | Multiple policies same policyholder | Policy count per holder | ### 2. Claimant Network Engine (`src/claimant_engine.py`) Staged accident rings and repeat claimants. | Signal | Method | |--------|---------| | Repeat claimant across incidents | ID frequency across claims | | Shared address per incident | Unique address count per claim | | High prior claims | Prior claim count threshold | | Many third parties | Third-party count per claim | | Repair shop concentration | Shop claim share per claimant group | ### 3. Repair & Quantum Engine (`src/repair_engine.py`) Infl …