Adaptive Intrusion Detection Mechanisms For Enhancing Security In Cloud-Hosted Big Data Systems

Abstract

The development and evaluation of a new Intrusion Detection System (IDS) using CNN and SVM is the main contribution of this paper. The hybrid approach using the feature extraction strengths of CNNs for very powerful invariant features is proposed to identify complex patterns that indicate possible intrusions in achieving automated network traffic data analysis. The model enhances detection with higher effectiveness across different attack vectors by incorporating the use of SVM as a classifier. Extensive experiments have been carried out to compare the performance of the proposed CNN+SVM model against traditional IDS methodologies. The results indicate a significant improvement in the detection rate, accuracies reaching a high of 98% with this hybrid model while keeping false positives to a minimum, whereas true positive cases are maximized. Other measures, such as precision, recall, and F1-score, further establish the fact that a model can compete by striving to balance sensitivity and specificity in order to handle one of the most critical challenges that cybersecurity faces. These findings demonstrate the potential benefit of deep learning models coupled with traditional machine learning techniques for intrusion detection systems. The contribution of this work to the extension in the literature on IDS, and above all, to some important practical implications in securing network infrastructures from the new, continuously evolving cyber threats, is important. Future studies will focus on model optimization and its application in different network scenarios.