Extended Framework for ML models in Streaming Anomaly Detection

This is a complimentary repository for the publication [Insert publication].

Fundamental Tasks

A streaming anomaly detection algorithm is formalized to consist of 4 fundamental tasks:

• Data representation
• Learning strategy
• Nonconformity score
• Anomaly score

Code Design

Every method for a fundamental task is implemented as a class according to an abstract class of that task. Thereby, a method includes one publisher and multiple subscribers. Instances of methods for different tasks are envisioned to be connected as Publisher <--> Subscriber according to the Observer pattern. In order to retain a low code complexity, instantiation and connection of fundamental task methods is done only in main.py.