| dc.description.abstract | The availability of time series streaming data has increased dramatically in recent scenarios.
During the previous decade, learning from real-time information has gained popularity. When
collecting significant information from data streams, machine learning has to deal with the shift
in data distribution. The changes in data distribution give rise to the appearance of hidden data
contexts that learning systems do not know about. This phenomenon, called concept drift,
deteriorates the classifier’s accuracy since the learning model classifies incoming instances
based on previous training data. Traditional classifiers struggle to identify patterns in nonstationary data distributions. The core missions of real-time classifiers must recognize concept
drift and react accordingly. The goal of this thesis report is to introduce widely used concept
drift detectors, along with their important features, strengths and weaknesses. At the same time,
this report also sets up experiments to evaluate the effectiveness of the calculations when
deployed in practice. | en_US |
