Financial fraud is rarely identifiable from a single transaction in isolation. In real-world systems, fraud is detected through patterns of behavior over time—sudden spending spikes, unexpected location changes, new devices combined with rapid transactions, and deviations from a user’s historical habits.

This competition challenges participants to go beyond traditional point-based classification and instead build models that understand how user behavior evolves across sequences of transactions. Each row in the dataset represents a transaction, but the true signal lies in the temporal context, historical baselines, and behavioral shifts associated with each user.

Rather than asking, “Is this transaction suspicious on its own?”, this challenge asks:
“Does this transaction break the story of the user’s past behavior?”

Participants are encouraged to engineer sequence-aware features, model temporal dependencies, and capture behavioral anomalies that reflect how modern financial institutions detect fraud at scale. The dataset is designed to simulate realistic user profiles, spending habits, and fraud patterns, making this competition both technically challenging and industry-relevant.