An Improved Link Prediction Algorithm for Financial Transaction Graphs Based on Structural Similarity and Firm Attributes

Accurate link prediction in financial transaction networks is essential for fraud detection, risk management, and financial auditing. This paper introduces an advanced hybrid algorithm that combines structural similarity metrics (e.g., Jaccard index, Adamic-Adar) with firm-specific attributes (e.g., industry, financial health, transaction history) to significantly improve prediction accuracy over traditional topology-only methods. Our methodology integrates data collection, feature extraction, and machine learning modeling (e.g., Random Forests, SVMs), demonstrating a 20–25% increase in accuracy and achieving precision and recall rates above 80% in empirical tests on real-world financial datasets. The algorithm’s scalability and interpretability enable actionable insights for identifying hidden transactional relationships, offering financial institutions a robust tool for proactive risk assessment, fraud mitigation, and strategic decision-making. By bridging structural and contextual analysis, this research advances graph-based machine learning in finance and sets a foundation for future innovations in dynamic transaction networks.