Comparative Analysis of Ensemble Based Learning and Hybrid Models for Improving DoS Attack Prediction Accuracy in Networking Environments


  • Gottapu Sankara Rao, P. Krishna Subbarao


DDoS assaults threaten computer networks and systems. These attacks flood the targeted system with traffic from many sources, disrupting service. Cyber security now requires real-time attack detection.. The current approach of detecting DDoS attacks is plagued by the issue of elevated percentages of false positives. Moreover, the classifiers employed in the current methodologies may lack the capability to comprehend the intricate patterns of the DDoS assault flow, resulting in diminished accuracy. This paper presents a refined method for identifying DDoS assaults by utilizing a classifier based on ensemble learning and a hybrid machine learning model. The ensemble-based Voting Classifier has the capability to combine many machine learning algorithms (SVC, Logistic Regression, Random Forest, and Naïve Bayes) in order to enhance classification accuracy. This makes it a superior choice for detecting DDoS attacks compared to a single machine learning-based classifier. The challenge of detecting Denial of Service (DoS) attacks in extensive sets of network traffic data is addressed by employing an ensemble of classifiers, known as the Hybrid method, which is designed to survive network attacks. The aim of this study is to construct a collection of classifiers that outperforms individual classifiers in terms of accuracy. The classifiers used in this experiment include SVC, Logistic Regression, Random Forest, and Naive Bayes. The suggested approach is compared against single classifiers using the metrics of accuracy, precision, recall, and F-measure. Outcome. The tests were conducted using Python 3.7.4 in Jupyter Notebook. The studies utilized the publicly available NSL-KDD dataset, which consists of network traffic data. The dataset was segmented into two classes: attack and regular network behavior traffic. The conducted experiments have confirmed the efficacy of the proposed methodology. The suggested Hybrid ML Model for network threats identification surpasses the performance of individual classifiers as well as the Ensemble Based Classifier. The Hybrid ML model attained a detection accuracy of 98.41%, whilst the Ensemble learning Method achieved a detection accuracy of 97%. The results imply that the proposed method has greater efficacy in terms of accuracy when compared to other individual classifiers such as SVM, Naïve Bayes, and Logistic regression and ensemble learning. In order to improve the analysis of network traffic attacks, investigations will be carried out on genuine big data sets.