Adaptive KDE for Real-Time Thresholding: Prioritized Queues for Financial Crime Investigation

TL;DR

This paper introduces an adaptive kernel density estimation method for real-time thresholding in non-stationary environments, specifically applied to financial crime detection, ensuring stable and semantically consistent risk score partitioning.

Contribution

It proposes a novel adaptive KDE approach that dynamically adjusts to changing score distributions for improved real-time decision thresholds.

Findings

Enhanced stability of thresholds over non-stationary data

Effective prioritization of risk scores in financial crime detection

Maintains semantic consistency in decision regions

Abstract

We study the problem of converting a continuous stream of risk scores into stable decision thresholds under non-stationary score distributions. This problem arises in a wide range of detection systems where scores must be partitioned into prioritized processing regions while preserving semantic consistency over time.