Influence Of Fractal Dimension Statistical Charachteristics On Quality Of Network Attacks Binary Classification

Abstract

It is proposed to increase abnormal behavior of network attacks binary classification efficiency by introducing additional informative features. Wide range of statistical characteristics of fractal dimension (FD) of the processed sequences was suggested to use as additional features. The effectiveness of the proposed method is shown by evaluating network attacks and normal traffic binary classification quality with machine learning algorithms in case of using the UNSW-NB15 database. Usage of FD statistical characteristics as additional informational features makes it possible to increase the efficiency of binary classification by an average of 10% for k-nearest neighbor and logistic regression algorithms. For the ""random forest"" and ""decision tree"" algorithms, the greatest effect of using additional FR parameters is observed in reducing the time spent on training and testing classification algorithms by more than 3.5 times.

Keywords

• classification
• fractal dimension
• hurst exponent
• attributes
• training
• testing
• metrics
• algorithms
• anomalies
• attacks
• unsw-nb15