Network Analysis and Text Mining for Identifying Money Laundering in Transaction Records

Abstract

Financial transactions have become increasingly complex, raising the potential risk of suspicious activities such as money laundering and fraudulent practices. This research addresses the problem of limited detection accuracy when using a single analytical approach and aims to propose an integrated framework combining graph analysis and text mining. The study adopts a quantitative design, utilizing transaction records with both numerical and descriptive attributes. Data preprocessing was conducted through cleaning, normalization, and feature extraction, followed by network analysis to build transaction graphs, measure centrality, and detect communities. In parallel, text mining was employed using keyword extraction, TF-IDF weighting, and clustering of transaction descriptions. The integration of these methods produced a hybrid classification model, which was then evaluated using accuracy, precision, recall, and F1-score metrics. The results show that the hybrid model significantly outperforms standalone approaches, reducing both false positives and false negatives while providing a more comprehensive detection system. The findings suggest that the combination of structural and semantic perspectives enhances the reliability of suspicious transaction identification. This study concludes that integrated analytical approaches can serve as a more effective tool for financial institutions and regulators in addressing fraudulent activities, while also offering potential for further refinement through deep learning integration and real-time processing in future work.