The exponential growth of digital infrastructures and the increasing sophistication of cyber-attacks necessitate the development of intelligent, adaptive, and real-time defense mechanisms. Traditional signature-based intrusion detection systems often fail to detect zero-day exploits and evolving attack patterns, making anomaly detection a critical component of modern cybersecurity. This research proposes an Adaptive Machine Learning Framework capable of detecting anomalies in real time by integrating streaming data analysis, dynamic feature selection, and continuous model optimization. The framework leverages a hybrid learning paradigm that combines supervised and unsupervised techniques—specifically, ensemble-based classification for known threats and clustering-based outlier detection for unknown patterns. A key innovation lies in the adaptive retraining module, which incrementally updates the model parameters in response to evolving network behaviors and attack signatures without requiring full retraining, thereby reducing computational overhead. The system architecture incorporates data preprocessing, feature engineering, adaptive model selection, and decision fusion layers to ensure high detection accuracy and minimal false positives. Real-world network traffic datasets, such as UNSW-NB15 and CIC-IDS2017, were used to validate the framework’s effectiveness. Experimental results demonstrate an average detection accuracy exceeding 98% with a significant improvement in detection latency compared to baseline methods. This approach shows strong potential for deployment in live cybersecurity environments, offering robust defense against both known and unknown threats. The proposed framework can be extended to support multi-modal data sources, enabling its integration into large-scale security information and event management (SIEM) systems for proactive threat mitigation.
Thangatharani et al. (2025) studied this question.