9 Anti-money laundering for Bitcoin transactions using graph attention network

 

This chapter covers

  • Understanding money laundering in Bitcoin transactions
  • Exploring the Elliptic Bitcoin graph dataset
  • Catching illicit Bitcoin transactions using traditional deep learning
  • Using Graph Convolutional Network (GCN) to detect illicit transactions
  • Improving performance using Graph Attention Network (GAT)

Money laundering is the act of hiding or whitewashing money gained from crimes by moving it through legal businesses or accounts to make it look clean. Why the term “laundering”, though? Al Capone, the infamous American gangster in the 1920s, is said to have coined the term, as he used to invest in laundromats to whitewash the cash he earned illegally. Restaurants, clubs, breweries, or even real estate are some other businesses used historically as black money cleansers. As shown in figure 9.1, money laundering is conducted in three steps – placement, layering, and integration.

Figure 9.1 Money laundering cycle, where illegally earned money is placed into the legal financial system via particular businesses such as laundromats, clubs, real estate, and so on. This money is then layered through a complex mesh of transactions into different financial entities, making it difficult to trace origins and identities associated with the laundered money. This money is then integrated into mainstream economy via clean investments into luxury assets, and different industrial and commercial entities.

9.1 Understanding the Elliptic Bitcoin dataset

9.1.1 Downloading the dataset with PyTorch Geometric

9.1.2 Analyze Bitcoin transaction statistics

9.1.3 Process the Elliptic Bitcoin dataset for deep learning

9.1.4 Visualize licit and illicit Bitcoin transactions

9.2 Using MLP to detect illicit Bitcoin transactions

9.2.1 Defining the MLP model architecture

9.2.2 Visualizing untrained model’s predictions on a 2D plot

9.2.3 Training and evaluating the MLP model

9.2.4 Visualizing trained model’s predictions on a 2-D plot

9.3 Detecting illicit Bitcoin transactions using GCN

9.3.1 Understanding the intuition behind graph learning

9.3.2 Building a GCN model to detect illicit transactions

9.3.3 Visualizing untrained GCN model’s predictions

9.3.4 Training and evaluating the GCN model

9.3.5 Visualizing the trained model’s output

9.4 Using GAT to improve illicit transaction detection performance

9.4.1 Understanding GAT model

9.4.2 Building a GAT model for anti-money laundering

9.4.3 Visualizing untrained GAT model’s predictions

9.4.4 Training and evaluating GAT model

9.4.5 Visualizing trained GAT model’s predictions

9.5 Summary