Answers are grounded in this paper and cite their sources.
Introduction. The introduction of machine learning methods allows automatic classification, analysis and filtering of malicious and unwanted mobile applications of network traffic, which may pose a threat to the integrity or availability of data, and unwanted ones a threat to confidentiality. Since the characteristics of such data change over time, this creates problems for accurately predicting classification models. The phenomenon when the characteristics and distribution of the data change, which leads to the need to update the model, is called "concept drift", and the adaptation of the model to new changes is called "concept drift adaptation". In such non-stationary environments, pre-trained machine learning models quickly become outdated and need to be continuously updated, leading to the need to detect concept drift. Given the non-stationary nature of the data, concept drift detection and adaptation is an important component of machine learning solutions that operate on data streams to maintain their integrity, validity, and confidence levels. Results: To solve such problems, methods based on mathematical statistics and machine learning technologies are now widely used, with the help of which even unknown malicious applications can be detected with a certain degree of probability. Based on a systematic review of the literature, the main groups of recent scientific publications on conceptual drift detectors are presented. Controlled (statistical, window and ensemble), semi-controlled, uncontrolled, ensemble methods for detecting concept drift are considered. The features of online drift detection based on windowing and batch processing are analyzed. A taxonomy of modern approaches to uncontrolled concept drift is proposed.
No takes yet. Share an insight, caveat, or question.
Sheluhin et al. (2023) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: