Graph-Powered Analytics and Machine Learning with TigerGraph

Chapter 12. Improving Fraud Detection

In an earlier chapter, we took on the problem of fraud detection by designing graph queries that looked for certain patterns of behavior that could be suspicious. This chapter will apply machine learning methods to improve fraud detection. Machine learning can help us via anomaly detection or by training the software to recognize fraud based on examples of known fraud cases. In both cases, graph-structured data is a valuable asset for sensing the unusual (anomalies) or for supplying data features (to build predictive models). No method is perfect, but machine learning can often detect patterns and anomalies that humans would miss. Conventional approaches only follow the rules that experts dictate. Using machine learning on graphs, we can detect patterns within the data that were not explicitly flagged as fraud cases, which makes it more adaptive to changing fraud tactics.

After completing this chapter, you should be able to:

• Deploy and use the TigerGraph Machine Learning Workbench

• Use graph-based features to enrich the feature vector of a dataset and then compare the model accuracies with and without the graph features

• Prepare data for and train a graph neural network for node prediction—in this case, fraud prediction

Goal: Improve Fraud Detection

Fraud is the use of deception for personal enrichment. Fraudsters might sabotage a system and its users, but in the end it is for personal gain. Examples of fraudulent activities are identity ...