Efficient Cloud IDS Framework Using Automated Feature Encoding and Multi-Model Evaluation

Abstract

This work provides an explanation for an efficient method for the detection of Distributed Denial-of-Service (DDoS) traffic on cloud networks, achieved through the development of an elucidated machine learning process. This process begins with an exploration and methodically thorough evaluation of the selected dataset, followed by extensive cleaning, encoding, scaling, and partitioning. Three respective classifications achieved using the Random Forest, Support Vector Machine (SVM), and the development of Logistic Regression algorithms, are performed on the partitioned data. The preliminary testing provides an accurate reading of 99.83%, along with the presence of high precision, recall, and F1-score values, substantiating the efficient potential of classical algorithms coupled with expert-level attention to detail, as well as complete and effective data preparation, for optimal functionality on cloud networks. In comparing the model to that of other contemporary research concerning efficient cloud network intrusion detection, it becomes immediately clear that the integration of the proposed efforts maintains accordance and, in some instances, beats the functionality of hybrid and Bayesian-based CNN-RNN architectures. This finding correlates well to the fact that, despite the crucial functionality and integral involvement of advanced classical and AI-driven architectures, expert-level attention to detail serves as the determining factor for advanced functionality and performance. This work illustrates the high degree of requisite attention necessary during the development and design process for an efficient cloud network intrusion technique, maintaining the inherent import and potential that offers the involvement and development of high-performing classical architectures.