CyberAdaptAI: A Dynamic Ensemble Learning Framework for Real-Time Cyberattack Detection Using AdaptEnsembleNet

Networked systems have been expanding rapidly, and there are cybersecurity challenges that require advanced Intrusion Detection Systems (IDS) to detect sophisticated and evolving threats. However, the more common traditional IDS approaches, including signature-based and classical machine learning methods, usually suffer from a significant drop in performance as they typically cannot adapt well to concept drift and data imbalance and cannot provide enough interpretability 6-9. In dynamic networks, these challenges prevent faster and accurate detection of new attacks or zero-day attacks. This article presents CyberAdaptAI, a novel hybrid adaptive ensemble learning framework that combines several base classifiers through an efficient drift detection scheme and adaptive weight rebalancing to overcome these limitations 38. It has also integrated explainability through SHAP-based interpretability, leading to actionable insights for security analysts. The general approach is to apply mini-batch processing of the streaming network data, dynamically tuning the classifier weights based on the most recent performance, concept detection using ADWIN, and a mechanism triggered by concept drift detection to train new models to maintain accuracy. CyberAdaptAI achieves up to 98.1% and 96.8% accuracy on benchmark datasets CIC-IDS2017 and UNSW-NB15, while outperforming state-of-the-art baselines empirically evaluated. Not only does the model recover quickly after encountering drift events, but it is also consistent and stable during batch-wise performance. Besides, cross-dataset evaluations substantiate its robustness and generalization abilities in a heterogeneous network scenario. The solution provided by CyberAdaptAI enables a practical and scalable approach to real-time intrusion detection in complex and evolving cyber environments, relying on adaptability, accuracy, and interpretability. By seamlessly enabling network behaviors of relevance and integrated with transparent decision-making, the framework adds novel support for security operations and threat mitigation, addressing critical gaps in existing IDS methodologies.