An Enhancing Network Security: A Stacked Ensemble Intrusion Detection System for Effective Threat Mitigation

• Published in: 2023 3rd International Conference on Innovative Mechanisms for Industry Applications (ICIMIA) pp. 1314 - 1321
• Main Authors: Vullam, Nagagopiraju, Roja, D, Rao, NagaMalleswara, Vellela, Sai Srinivas, Vuyyuru, Lakshma Reddy, Kumar, K Kiran
• Format: Conference Proceeding
• Language: English
• Published: IEEE 21.12.2023
• Subjects: Decision tree; Decision trees; Intrusion detection; Intrusion detection system (IDS); k-nearest-neighbour; Machine learning algorithms; random forest; Stacking; Telecommunication traffic; Testing; Training
• DOI: 10.1109/ICIMIA60377.2023.10426091

Modern cyber security relies heavily on intrusion detection in network traffic to quickly identify and mitigate security threats. Conventional intrusion detection systems frequently depend on lone, stand-alone models that could find it difficult to change with the ways that network attacks are evolving. In order to overcome this difficulty, this research study provides an ensemble-based strategy that improves threat detection effectiveness and accuracy by utilizing the strength of several intrusion detection models. Increasing security measures against data breaches and network invasions is essential due to the ever-growing usage of the Internet and networks. As intrusions are frequently hidden within of valid network packets, firewalls have a difficult time identifying and stopping them. Furthermore, the majority of network monitoring systems and algorithms find it increasingly difficult to handle the sheer volume of network traffic. Various intrusion detection strategies have been proposed in response to these issues, with machine learning techniques emerging as a possible route for handling these situations. This paper introduces an Intrusion Detection System (IDS) that leverages stacking ensemble learning. The core ensemble comprises three fundamental machine learning models: k-nearest-neighbours, Decision Tree, and Random Forest. To enhance classification performance, the proposed system combines a total of seven machine learning algorithms with preprocessing methods. Stacking ensembles improve system performance greatly by combining the results of these basic models with a meta-model represented by the Logistic Regression algorithm. The UNSW-NB15 dataset is used to assess the IDS's efficacy. With an astounding 96.16% accuracy rate in the training phase and an even greater 97.95% accuracy rate in the testing phase, the suggested IDS performs remarkably well. Additionally, the precision scores are remarkable, with training scores of 97.78% and testing scores of 98.40%. These outcomes highlight the system's capacity to detect and stop network intrusions, showing considerable gains in a number of assessment parameters.