Data Science–Enabled Anomaly Detection in Financial Transactions Using Autoencoders and Risk Evaluation Mechanisms

Main Article Content

Naga Charan Nandigama

Abstract

Financial institutions increasingly face sophisticated fraudulent activities that challenge the capabilities of traditional rule-based detection mechanisms. This study proposes a data science–enabled anomaly detection model that integrates autoencoder-based deep learning with a structured risk evaluation mechanism to identify suspicious financial transactions with high accuracy. The framework leverages transactional history stored in a data warehousing environment to generate enriched behavioral features and reconstruct patterns using an autoencoder architecture. Transactions producing high reconstruction errors are assigned anomaly scores, which are further refined through a complementary risk assessment layer that incorporates user behavior, transaction context, and domaindriven risk rules. The hybrid design enhances detection of both known and previously unseen anomalies while minimizing false positives. Experimental results demonstrate strong performance across multiple datasets, indicating that the combination of deep learning and risk evaluation significantly improves fraud detection capabilities. The proposed model offers a scalable and adaptive solution suitable for real-time financial monitoring systems.

Article Details

Section
Articles