Author : Mani Gopalsamy

The capacity to detect facts or observations that differ from what is normally thought of by domain experts is crucial for many contemporary applications. These outliers may be located with the use of anomaly detection, and the system can subsequently implement the required adjustments. This study presents a scalable anomaly detection framework for network traffic analysis in cybersecurity using advanced machine learning. Approaches. Leveraging the NSL-KDD dataset. Before model building, Recursive Feature Elimination (RFE) identifies the most relevant features for classification. Machine learning models—DNN, KNN, RF, and NB—are employed and evaluated using F1-score, recall, accuracy, and precision, with a confusion matrix to assess performance. Results show RF achieves the highest accuracy (99.81%), precision (99.89%), and recall (99.90%), followed closely by KNN. Generally speaking, both NB and DNN perform worse since their metrics are lower. The results reveal the enhanced performance of RF and KNN in terms of identifying and categorising anomalous behaviours in the network traffic and, therefore, providing a viable solution to augment existing real-time cybersecurity systems.

Keywords: Anomaly detection, Network traffic analysis, Cybersecurity, Intrusion detection systems (IDS), machine learning.