A SUSPICIOUS FINANCIAL TRANSACTIONS DETECTION MODEL USING AUTOENCODER AND RISK BASED APPROACH

Abstract

Financial institutions process millions of transactions daily, making the identification of suspicious financial activities a critical component of fraud prevention and anti-money laundering (AML) efforts. Traditional rule-based detection systems often struggle to identify complex and evolving fraudulent patterns due to the increasing volume, velocity, and diversity of financial transactions. This paper presents a suspicious financial transaction detection model that integrates autoencoder-based anomaly detection with a risk-based assessment approach. The proposed framework utilizes autoencoders to learn normal transaction behaviors and identify anomalous activities by measuring reconstruction errors. Transactions exhibiting significant deviations from learned patterns are flagged as potentially suspicious. In addition, a risk-based evaluation mechanism is incorporated to assess transaction risk levels based on factors such as transaction amount, frequency, customer behavior, geographic location, and historical activity patterns. The combination of deep learning-based anomaly detection and risk assessment enhances detection accuracy and reduces false positive rates. Experimental analysis demonstrates that the proposed model effectively identifies suspicious financial transactions, improves fraud detection capabilities, and supports proactive financial risk management. The developed framework provides an intelligent, scalable, and adaptive solution for strengthening financial security and regulatory compliance.