| Abstract | Identification of rare events is an important component in intrusion and fraud detection, system monitoring and event detection. Anomalies can represent problematic situations where early, accurate and actionable insights are critical to make situational assessments in the event of unexpected conditions. For many problems, the state of the art in machine learning is batch learning. However, training models in a batch context can represent an opportunity cost. For instance, in applications such as fraud detection, the time and investment spent on training batch models, with increasingly larger data sets and complex infrastructure, could be spent detecting aberrant behavior. Online anomaly detection algorithms offer rapid access to useful insights with fewer computing capacity requirements and often need to be integrated with existing or legacy data streams in the enterprise. This work introduces a proof of concept integration of two online streaming anomaly detection algorithms available in the scikit-multiflow and River frameworks with the popular Kafka event streaming platform. |
|---|
