HPDDoS: A HyperParameter Model for Detection of Multiclass DDoS Attacks

Authors

  • Ms. Vimal Gaur, Dr. Rajneesh Kumar

Keywords:

ANOVA, Chi-Square, DDoS, Extra Tree, Hyper Parameter Optimization, Mutual Information

Abstract

In the modern era, network attacks are increasing at a breakneck pace. However, providing proper security measures to mitigate the different attacks is still a significant challenge. Distributed denial of service attacks are most dangerous as they bring whole network down. Recent work has relied on black-box optimization with hyperopt and machine learning models for handling DDoS attacks. In this paper, a hyperparameter optimization model to detect DDoS attacks has been proposed to enhance the efficiency of DDoS detection in terms of accuracy and training time. We demonstrate the feasibility of this model by comparing the performance measures of machine learning algorithms with and without feature selection methods. The feature selection methods (Chi-Square, Extra Tree, ANOVA, and Mutual Information) have been applied for reducing features and machine learning algorithms (Random Forest, Decision Tree, XGBoost, KNN, SNN, and DNN) for classification. Results show that the Random Forest classifier combined with Mutual Information feature selection achieves 96.77% as maximum accuracy in 750.94 seconds of training time. Hyperparameter tuning of the random forest classifier raises the accuracy value to 96.81%. Later, with a 95% confidence level, the confidence intervals of the default and hyperparameter values of Random Forest were calculated to be 98.58% and 98.78%, respectively. The results show that the hyperparameter model shows a 2.01% improvement in accuracy with multiclass data.

Downloads

Published

2022-08-08