OPTIMIZATION OF LARGE TRANSACTION GRAPH VISUALIZATION FOR FRAUD PATTERN DETECTION USING FORCE-DIRECTED LAYOUT ALGORITHMS

Abstract

This study investigates the problem of visualizing large transaction graphs in the context of fraud detection in financial systems. The relevance of the study is driven by the rapid growth of transactional data volumes and the increasing complexity of their structural organization, which complicates both automated analysis and expert interpretation. It is shown that classical force-directed layout algorithms, when applied to graphs with hundreds of thousands of nodes and edges, fail to provide sufficient performance, making their use in interactive visual analytics systems impractical.

The study justifies the decomposition of the problem into two levels: graph preprocessing and optimization of force computation during layout. An approach is proposed that combines the filtering of low-significance structural elements with a topology-aware approximation of repulsive forces. The key idea is to replace pairwise node interactions with interactions between nodes and the centers of mass of connected components, which significantly reduces computational complexity without critical loss of visualization quality.

An experimental study is conducted on the synthetic SAML-D dataset, which models transaction processes with embedded money laundering scenarios. The evaluation considers metrics such as per-iteration time, the number of iterations required to reach a target quality level (based on neighborhood preservation), and visual layout quality. The results demonstrate a substantial reduction in iteration time and total number of iterations, with only minor degradation in visualization accuracy.

It is shown that the proposed approach enables interactive exploration of large graphs while preserving key structural patterns, such as hubs and transaction chains, which are critical for financial monitoring tasks. The results can be applied in the development of visual analytics systems to support decision-making in the AML domain.

Keywords: graph visualization, force-directed algorithms, Fruchterman–Reingold, fraudulent transactions, AML, transaction networks.