Towards Federated Graph Learning for Collaborative Financial Crimes Detection

Abstract: Financial crime is a large and growing problem, in some way touching almost every financial institution. Financial institutions are the front line in the war against financial crime and accordingly, must devote substantial human and technology resources to this effort. Current processes to detect financial misconduct have limitations in their ability to effectively differentiate between malicious behavior and ordinary financial activity. These limitations tend to result in gross over-reporting of suspicious activity that necessitate time-intensive and costly manual review. Advances in technology used in this domain, including machine learning based approaches, can improve upon the effectiveness of financial institutions' existing processes, however, a key challenge that most financial institutions continue to face is that they address financial crimes in isolation without any insight from other firms. Where financial institutions address financial crimes through the lens of their own firm, perpetrators may devise sophisticated strategies that may span across institutions and geographies. Financial institutions continue to work relentlessly to advance their capabilities, forming partnerships across institutions to share insights, patterns and capabilities. These public-private partnerships are subject to stringent regulatory and data privacy requirements, thereby making it difficult to rely on traditional technology solutions. In this paper, we propose a methodology to share key information across institutions by using a federated graph learning platform that enables us to build more accurate machine learning models by leveraging federated learning and also graph learning approaches. We demonstrated that our federated model outperforms local model by 20% with the UK FCA TechSprint data set. This new platform opens up a door to efficiently detecting global money laundering activity.

• The paper presents a federated graph learning platform enabling collaborative detection of financial crimes across institutions.
• It reduces false positives by 20-30% through integrating local and global features with advanced graph analytics.
• The approach ensures regulatory compliance via privacy-preserving federated training and scalable detection improvements.

Towards Federated Graph Learning for Collaborative Financial Crimes Detection

Introduction

The paper "Towards Federated Graph Learning for Collaborative Financial Crimes Detection" (1909.12946) discusses a novel approach for improving the detection of financial crimes such as money laundering, fraud, and various forms of theft. The central theme revolves around augmenting traditional AML systems within financial institutions using a federated graph learning framework. This framework allows the sharing of insights and patterns without exchanging sensitive raw data, adhering to stringent privacy and regulatory requirements.

Financial Crimes and Current Detection Challenges

Financial crimes present expansive challenges due to their complexity and the significant resources required for effective detection and mitigation. Traditional systems are hindered by over-reporting false positives, necessitating costly manual reviews. The paper identifies a critical limitation: financial institutions operate in isolation without insights from other firms, limiting the detection of sophisticated techniques perpetrated by criminals across multiple geographies and institutions.

Machine Learning and Graph Learning Integration

The integration of machine learning, specifically graph-based approaches, is a promising method that current detection strategies can utilize. By constructing party-to-party relationships and leveraging graph-based features, machine learning models improve the efficiency of transaction monitoring systems. This methodology effectively reduces false positives by 20-30% through improved anomaly detection, leveraging network embeddings, and traditional graph algorithms.

Federated Graph Learning Platform

The core contribution of the paper lies in its development of a federated graph learning platform. This platform allows institutions to share essential features extracted from their local data, such as customer demographics and transaction metrics, and also global graph features that are computed using a joint representation of interactions across institutions. The federated learning paradigm ensures privacy-preserving model training and enhances the detection capabilities beyond isolated models.

Local and Global Feature Computation

Local features derived from customer data and transaction behaviors form the initial input for models. Global features, providing contextual data across institutions, enhance detection capabilities. The paper details graph analytics to derive global features, including metrics from community detection algorithms and multi-hop egonets, emphasizing the crucial insights gained from these enriched data structures.

Figure 1: Relation between customers and related parties.

Experimental Evaluation

The framework was evaluated using data from the UK FCA TechSprint event dataset, comprising profiles and transactions from multiple UK-based institutions. The novel use of federated graph learning led to a 20% improvement over local models. Significantly, the approach maintained data privacy, adhering to financial regulations and competitive neutrality.

Figure 2: Conditional probability of a customer's fincrime exit marker being flagged as a function of a feature. Left: the feature is the number of SAR-flagged customers in the same connected component. Right: the feature is the number of nodes in the connected component.

Implementation Considerations

Implementing such a federated graph learning system requires addressing several technical considerations, including the secure computation necessary to protect sensitive information. Furthermore, ensuring system scalability to work with vast amounts of data from real-world applications is crucial. The system should enable decentralized training with secure communication protocols to reconcile global model parameters without compromising institutional data privacy.

Conclusions and Future Directions

The federated graph learning platform presents significant potential for enhancing global financial crime detection mechanisms. Future work involves real-world piloting, exploring roles for different industry stakeholders in deploying these techniques effectively, and integrating advanced secure computation frameworks for entity resolution processes. This work substantiates a foundation for collaborative, privacy-preserving financial crime detection across multiple financial institutions globally.