<![CDATA[In the modern network environment characterized by highly heterogeneous network architectures, the security landscape is dominated by a persistent and burgeoning threat scenario of extremely high complexity. Security systems based on signature detection are not efficient at protecting against new threats, and they are vulnerable to evasion strategies.In this research, we propose a novel machine-learning network intrusion detection system (NIDS) designed for multiclass network attack classification based on a weighted Random Forest algorithm. The proposed model was tested experimentally on the well-known benchmark dataset KDDCUP99 that contained 494,021 network connections. We designed the preprocessing pipeline based on macro-class level categorical label mapping, ordinal attribute mapping, and Min-Max numerical normalization.The evaluation was performed using the validation subset that was kept separate (i.e., an out-of-sample test) from the training process and consisted of 148,207 records. The proposed model achieved an overall accuracy of 99.97% in multiclass classification. We focus on discussing the severe class imbalance, and the experimental results indicate that our method is capable of achieving satisfactory predictive performance for minority classes, as the proposed classifier maintains a precision score as high as 90.91% for a rare attack class User-to-Root (U2R). Finally, we provide a complete comparison of the proposed approach with a set of state-of-the-art intrusion detection baselines.]]>
Copyrights © 2026
