A Novel Efficient Hybrid Machine Learning Architecture for Classification and Detection of Network in Cybersecurity Frameworks

Background: Security for industrial and robotic systems depends on network-based intrusion detection systems (NIDS) due to evolving cyber threats. Traditional intrusion detection systems (IDS) often fail in these environments, necessitating advanced machine learning (ML) approaches. Methods: This study applies ML techniques to the UNSW-NB15 dataset for intrusion detection. Data preprocessing entails the optimisation of dataset size, the elimination of redundancies, and the management of absent values. Random Forest is employed to identify critical features, while SMOTE is implemented to mitigate class imbalance. Hybrid Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM)—is implemented and evaluated, also compared with existing KNN and DNN. Results: The Hybrid CNN-BiLSTM model outperforms other models, achieving 96.83% accuracy. Performance metrics, including F1-score, recall, precision, and accuracy, confirm its effectiveness. Visualisations of training-validation accuracy, loss, confusion matrices, and ROC curves validate model robustness. Conclusion: The Hybrid CNN-BiLSTM model exhibits exceptional performance, which helps to improve cybersecurity in industrial environments by facilitating the deployment of NIDS.