Adaptive Financial Fraud Detection in Imbalanced Data with Time-Varying Poisson Processes

TL;DR

This paper introduces a novel approach for financial fraud detection using Poisson processes, effectively handling imbalanced datasets and improving prediction accuracy over baseline methods.

Contribution

It develops a new methodology employing homogeneous and non-homogeneous Poisson processes for fraud detection in imbalanced financial data, enhancing predictive power.

Findings

Better prediction accuracy than baseline methods

Effective in highly imbalanced datasets

Applicable to real financial datasets

Abstract

This paper discusses financial fraud detection in imbalanced dataset using homogeneous and non-homogeneous Poisson processes. The probability of predicting fraud on the financial transaction is derived. Applying our methodology to the financial dataset shows a better predicting power than a baseline approach, especially in the case of higher imbalanced data.