Temporal Contrastive Transformer for Financial Crime Detection: Self-Supervised Sequence Embeddings via Predictive Contrastive Coding

In three linesTemporal Contrastive Transformer (TCT): self-supervised representation learning framework for financial fraud detection via transaction sequence embeddings. AUC 0.8644 standalone, 0.9245 combined with engineered features. Captures temporal structure but no additive gain over baseline.

Summary generated by Claude — human-verified