Graph-Aware Machine Learning Framework for Anti-Money Laundering in Cryptocurrency Transactions Using Hybrid Classification Models

Abstract

The rapid proliferation of cryptocurrency transactions has introduced unprecedented challenges in financial regulation, particularly in detecting illicit activities such as money laundering. Traditional Anti-Money Laundering (AML) systems are largely rule-based and struggle to adapt to the dynamic, decentralized, and pseudonymous nature of blockchain ecosystems. This research proposes a graph-aware machine learning framework that integrates conventional classification algorithms with deep learning techniques to enhance the detection of suspicious cryptocurrency transactions.The proposed system leverages transaction-level data, including features such as transaction amount and sender location, and applies preprocessing techniques such as label encoding, normalization, and Synthetic Minority Over-sampling Technique (SMOTE) to address data imbalance. Multiple machine learning algorithms, including Random Forest, Support Vector Machine (SVM), Naïve Bayes, Logistic Regression, and Decision Tree, are employed to classify transactions into legitimate or laundering categories. Additionally, a Convolutional Neural Network (CNN)-based model is introduced to capture non-linear feature interactions and improve classification performance.A key contribution of this work lies in its conceptual alignment with Graph Neural Networks (GNNs), where transaction relationships are implicitly modeled through structured feature learning. Although the implementation primarily utilizes tabular data, the framework is designed to extend toward graph-based representations for future scalability in blockchain analytics.The system is implemented using a Django-based web interface that enables real-time prediction and visualization of model performance. Evaluation metrics such as accuracy, precision, recall, and F1-score are computed to assess model effectiveness. Experimental results demonstrate that ensemble methods, particularly Random Forest, outperform other algorithms in terms of classification accuracy, while the CNN model shows competitive performance in capturing complex data patterns.The findings indicate that combining traditional machine learning techniques with deep learning models significantly enhances the robustness of AML systems. The integration of SMOTE improves minority class detection, addressing the critical issue of imbalanced datasets in fraud detection scenarios. Furthermore, the system provides an interpretable and scalable approach suitable for real-world financial monitoring applications.