Binary Classification of Fractal Time Series by Machine Learning Methods

Abstract

The paper considers the binary classification of time series based on their fractal properties by machine learning. This approach is applied to the realizations of normal and attacked network traffic, which allows to detect DDoS-attacks. A comparative analysis of the results of the classification by the random forest and neural network - fully connected multi-layer perceptron is carried out. The statistical, fractal and recurrence characteristics calculated from each time series were used as features for classification. The analysis showed that both methods provide highly accurate of classification and can be used to detect attacks in intrusion detection systems.