This document discusses e-commerce transaction fraud detection through machine learning. It proposes an approach using linear clustering of transaction data and features to create a hypergraph through entropy estimation and frequent itemset mining. The hypergraph is formed using Neo4j and analyzed to determine connections between features. Machine learning techniques like artificial neural networks are then used for classification to detect fraudulent transactions. The approach aims to improve on existing static rule-based fraud detection methods which criminals have learned to circumvent. Prior research on using support vector machines, graph computing, and semi-supervised/supervised techniques for fraud detection is also reviewed.