Automated Insurance Claims Fraud Screening Using Distance-Based Outlier Detection

Property and Casualty (P&C) insurance fraud inflicts severe financial strain on global
underwriting portfolios, driving up loss ratios, inflating policyholder premiums, and distorting
actuarial pricing models. Traditional claims screening relies on rule-based heuristic triggers and
manual Special Investigation Unit (SIU) referrals, which suffer from high false-positive rates,
administrative latency, and an inability to detect complex organized fraud syndicates. This paper
establishes an automated, real-time insurance claims fraud screening framework powered by
multi-metric distance-based outlier detection algorithms, including Mahalanobis Distance, $k$
Nearest Neighbors ($k$-NN) spatial density, and Local Outlier Factor (LOF). Grounded in non
linear risk parameterization methodologies and enterprise risk management architectures, our
model computes non-parametric spatial distance scores across high-dimensional claim feature
manifolds. Evaluated on an empirical dataset of 185,000 auto and property claims processed across 12 insurance carriers between 2018 and 2024, our multi-metric distance framework achieved an Area Under the Precision-Recall Curve (PR-AUC) of 0.884, a False Positive Rate (FPR) of 0.14%, and reduced SIU investigation triage latency from 18.5 days to 2.4 minutes. We present an exhaustive quantitative synthesis incorporating 10 data tables, 10 behavior graphs, and 20 visually distinct chart architectures detailing claim distance distributions, loss ratio optimizations, and spatial cluster dynamics.