Intelligent Machine Learning Approaches for Data-Driven Cybersecurity and Advanced Protection

Abstract

his paper examines the usage of cutting-edge machine learning (ML) techniques for enhancing data-centric cybersecurity, with a focus on detection, classification, and anomaly identification in different attack scenarios. In three case studies—IoT intrusion detection via convolutional neural networks (CNNs), ransomware detection with random forest classifiers, and unsupervised anomaly detection via the CAMLPAD framework—the work demonstrates how domain-specific ML models can address specialized threat environments. The CNN-based IoT intrusion model achieved 99.2% accuracy, 98.8% precision, and 99.0% F1-score across the UNSW-NB15 dataset, significantly better than the traditional statistical baselines. The random forest ransomware detection system achieved 98.5% accuracy, 97.9% recall, and area under the ROC curve (AUC) 0.995, showing robustness in distinguishing malicious and legitimate encryption activity. CAMLPAD identified 94.7% of anomalies with less than a 3% false positive rate and successfully identified zero-day attacks in real time without any labelled training data. . Comparative analysis reveals that supervised methods excel in well-characterised environments, while unsupervised models are indispensable for novel threat discovery. The study also addresses model explainability, adversarial robustness, and mitigation of human error, recommending an adaptive, multi-layered, and interpretable ML-driven cybersecurity architecture that combines continuous retraining, adversarial hardening, and human oversight to sustain high-performance cyber defence.