A Performance Analysis of ML-Based Intrusion Detection Systems in Cloud Environments

Intrusion Detection Systems (IDS) are important for protecting cloud environments against emerging cyber threats. This paper introduces AI-SCAN (artificial intelligence-driven scalable convolutional network for anomaly detection in cloud networks), a deep learning IDS that utilizes a Convolutional Neural Network (CNN) architecture to achieve better threat detection with better scalability, flexibility, and low false positives. The proposed system overcomes key challenges of dataset bias, external validation, and class imbalance to provide robust performance in dynamic cloud networks. To reduce dataset bias, we examine model performance on a variety of attack types and assess its efficacy with external validation on separate datasets outside the CSE-CICIDS2018 benchmark. Our solution combines SMOTE (synthetic minority oversampling technique)-based data augmentation and class weighting strategies to counteract minority attack classes, promoting model generalization. Hyperparameter tuning and feature selection also improve AI-SCAN’s efficiency, reducing computational overhead without sacrificing high detection accuracy. Empirical observations indicate 97.5% accuracy, 96.5% precision, and 95.0% recall, higher than conventional ML-based IDS implementations. AI-SCAN’s novel cyber threat detection with low false positives supports its applicability in real-time cloud deployment. The current study conducts a comparative analysis among conventional machine learning (ML), ensemble learning, and deep learning-based IDS models and positions AISCAN as a robust, scalable, and fault-tolerant cybersecurity measure.