Authors - Mohmed Umar, Jeevakala Siva Rama Krishna Abstract - In the era of complete digital connectivity, it is the need of the hour to keep the networks safe from a wide range of cyberattacks. Traditional Network Intrusion Detection Systems (NIDS) rely mainly on signature-based approaches; though highly efficient in identifying known threats, they suffer from weaknesses in discovering new and developing attacks, such as zero-day vulnerabilities. This results in higher false positives and lower detection efficiency. We present a novel NIDS based on the ensemble methods in machine learning, namely Random Forest and Bagging Classifiers, with which we may promise detection accuracy at the cost of a reduced level of false alarms. We conduct extensive evaluations based on systematic data preprocessing, feature selection, and model training against benchmark datasets like KDD Cup 99 and NSL-KDD. The system being considered achieves a detection accuracy of 99.81%, along with an F1 score of 99.82% and an AUC score of 99.81%, thus significantly surpassing the performance from traditional approaches. These results show the aptness of machine learning methodologies in enhancing network security, as it makes for a flexible and scalable solution suited for real-time deployment in extensive environments. Future work will focus on further developing the scalability of the system and minimizing latency to ensure seamless real-time operation.
Open Zoom
