Comparative analysis of anomaly recognition methods in real time

Abstract

The article discusses modern classes of algorithms used to detect anomalies in data streams: sliding
window algorithm, metric algorithms, predictive-based algorithms, and algorithms based on hidden
Markov models. During the research, it was possible to determine functional and efficiency criteria
for assessing the class of algorithms and then comparing it with other considered classes. In addition,
for each class of methods, strengths and weaknesses are given, the scope is described, and a generalized
example of implementation in the form of pseudo code is given. The use of this approach makes
it possible to cover entire groups of algorithms without reference to a specific implementation. The
conclusions obtained as a result of the research can be applied solving problems of optimizing the
process of detecting anomalies or increasing the efficiency of applied solutions used in these scenarios.
The resulting calculations allow further development and optimization of methods in this area
for unlabeled fixed data sets.