Fraud Detection with Apache Flink on Ververica

This repository contains sample code demonstrating Complex Event Processing (CEP) with Apache Flink on streaming financial transactions with Ververica.

The code showcases how to detect:

✅ High-Value Consecutive Transactions

✅ Multiple Rapid Transactions

✅ Location-Based Suspicious Activity

Overview

Financial transactions are generated using Java Faker, which provides randomized user data such as names, merchants, amounts, and locations. The transactions then flow through a Flink data stream to a series of CEP patterns:

• High-Value Consecutive Transactions
  Pattern of two consecutive transactions exceeding a certain threshold (e.g., >900) within 30 seconds.

• Multiple Rapid Transactions
  Pattern of three transactions within a short time window (e.g., 10 seconds).

• Location-Based Suspicious Activity
  Pattern of two transactions in widely different geographic locations (e.g., more than 500 km apart) within 1 minute.

By using keyBy(userId), we ensure these patterns apply per user, so any sequence that matches comes from a single user.

Sample Data Generation

• Transaction.java
  Defines the data model for financial transactions, including user ID, timestamp, amount, merchant, and location details.

• DataGenerator.java
  Uses javafaker to create realistic synthetic transaction data with random amounts, currency codes, merchants, and geolocation.

• TxnProducer.java Generates transactions and sends them to a Kafka topic. Make sure to fill in the required properties for bootstrap.servers and sasl.jaas.config into the AppConfig.java file.

Deployment

Run mvn clean package to create a jar file

On the Artifacts tab upload the generated jar file

Then navigate to the Deployments tab, click new deployment and put the required fields.

Finally click start and after a while you will see your job running.

Sample Output