An Empirical Survey of Fully Unsupervised Drift Detection Algorithms for Data Streams

This paper presents a comprehensive benchmark and survey of fully unsupervised concept drift detectors (UCDD) designed to identify and adapt to concept drift in real-world data streams. Concept drift refers to the phenomenon where the statistical properties of a data stream change over time, leading to the deterioration of model accuracy if not detected and adjusted. The study reviews the state of the art in UCDDs, evaluates their performance on various real-world datasets, and identifies challenges and open research areas in the field. Through empirical experiments and a systematic review of existing methods, we highlight key factors influencing the performance of these detectors in unsupervised environments.