Feature Engineering

Using ksqlDB / Kafka Streams, the platform continuously computes:

Velocity Features

• Transactions per wallet per second/minute

• Burst detection (sudden spikes)

Value Features

• Average vs current transaction amount

• Deviation from wallet baseline

Graph Features

• Number of unique counterparties

• New vs known address interactions

Temporal Features

• Wallet age

• Time-of-day anomalies

These features are windowed, e.g.:

• 10 seconds

• 1 minute

• 5 minutes

This is where Confluent is essential.