Balanced-BiEGCN: A Bidirectional EvolveGCN with a Class-Balanced Learning Network for Dynamic Anomaly Detection in...

Area of Science:

• Computer Science
• Financial Technology
• Network Security

Background:

• Bitcoin transaction anomaly detection is crucial for financial market stability.
• Current dynamic graph models face challenges in capturing long-range temporal dependencies and handling class imbalance in transaction data.
• Scarcity of abnormal samples complicates effective anomaly detection.

Purpose of the Study:

• To propose a novel approach, Balanced-BiEGCN, for enhanced Bitcoin transaction anomaly detection.
• To address limitations in capturing long-range temporal dependencies and class imbalance in existing methods.
• To improve the accuracy and robustness of anomaly detection in dynamic transaction networks.

Main Methods:

• Developed Bidirectional EvolveGCN (Bi-EvolveGCN) for enhanced capture of long-range temporal dependencies.
• Integrated a Sample Class Transformation (CSCT) classifier to generate difficult-to-distinguish abnormal samples, addressing class imbalance.
• Utilized adjacency distance adaptive loss and symmetric space adjustment loss functions to guide sample generation and optimize spatial distribution.

Main Results:

• The Balanced-BiEGCN model demonstrated superior performance in anomaly detection compared to existing baseline methods.
• Experimental results on the Elliptic dataset validated the effectiveness of the proposed approach.
• The bidirectional temporal feature fusion and class-balanced learning significantly improved detection capabilities.

Conclusions:

• Balanced-BiEGCN offers a significant advancement in Bitcoin transaction anomaly detection.
• The model's ability to handle dynamic patterns and class imbalance makes it a valuable tool for financial market stability.
• Future work could explore further refinements in temporal dependency modeling and sample generation techniques.