A Machine Learning‐Driven Cloud Intrusion Detection: Impact of Feature Engineering and Dimensionality Reduction on Classification and Optimization

Abstract

Cloud computing has emerged as one of the most rapidly advancing domains in modern information technology, enabling scalable and efficient resource utilization across diverse industries. However, with this expansion comes serious cyberattacks which can disrupt critical sectors, such as healthcare, telecommunications, and finance, making significant challenges to intrusion detection systems. To address these challenges, this paper presents a robust intrusion detection framework designed to enhance security in cloud computing environments. The contributions of this work begins with the selection of a benchmark dataset representative of real cloud‐based network traffic, followed by extraction of new critical features to identify the most relevant attributes for model development. We then leverage Principal Component Analysis (PCA) for feature reduction to enhance model efficiency. To detect intrusions, we employ multiple Non‐parametric and Parametric Learning Models and systematically evaluate their performance under four configurations: feature extraction with and without PCA, and PCA with and without feature extraction. Furthermore, to improve classification performance, we conduct hyperparameter optimization on the best performing algorithm. Finally, we conduct a comprehensive comparison with state‐of‐the‐art intrusion detection techniques, demonstrating the efficacy of our approach. Experimental results demonstrate that our optimized Multilayer Perceptron (MLP) model achieves the highest detection accuracy. This work not only advances intrusion detection in cloud networks but also provides insights into feature engineering and algorithmic optimization for improved cybersecurity.