Integration of multiple machine learning algorithms to detect unbalanced network traffic

Digital infrastructure networks use Network Intrusion Detection Systems (NIDS) security guards to protect against cyber threats. The detection process becomes difficult since networks transmit more normal traffic than malicious traffic. Resolving class imbalance problems leads to an improved deep learning NIDS framework which appears in this work through DSSSTE implementation for hard sample generation synthesis. DSSSTE establishes an optimized training procedure through which the data examples are segmented between simple and difficult portions while uniting malicious traffic oversampling methods with normal traffic under-sampling techniques to develop balanced datasets. The processed data increases model training results when used for analytical purposes. An evaluation of the proposed model took place through AUC along with accuracy, precision, recall, FPR and FNR when compared to Bi-LSTM, RF and SVM, LR and FNR when compared to LSTM, AlexNet, Mini-VGGNe. Security threat detection performance improves better than contemporary methods while producing lower false positive detections through the proposed system. This proposed research creates a reliable network security implementation for effective intrusion detection in systems operating with unbalanced conditions. Bi-LSTM achieves 99.0% accuracy, 92.0% F1-score, and 0.99 AUC, statistically outperforming all baseline models (p < 0.01).