Deep Learning-Based Network Intrusion Detection for Industrial IOT: Hybrid CNN-LSTM Architecture for Real-Time Threat Analysis

The idea of IIoT had grown at a very faster pace and it brought huge cyber security risks, especially to cyber- attacks. To mitigate the above-discussed threats, the convolutional neural network (CNN) and Long Short-Term Memory (LSTM) network Intrusion Detection System (IDS) model is introduced. This model improved the detection accuracy because the network traffic data from the IIoT network has both spatial and temporal dependencies. To test the proposed methods, we used the UNSW-NB15 dataset comprising of 175,341 records 49 features applied binary classification (normal and anomalous traffic) and multiple classification (9 classes of attacks). It started with feature scaling, missing data management and finally categorical data transformation. CNN layers capture spatial feature and LSTM layers capture the temporal feature. We employed the optimization algorithm Adam to train the model, and dropout layers were used to minimize over fitting. I used evaluation measures for evaluating effectiveness. Integration of CNN and LSTM models into a single architecture was shown to give much higher accuracy compared to the models where only one of the two was used; the accuracy for the binary classification came out to be 97% while for multi classification the accuracy was 99%. Compared with the previous results, false positive rates were much lowered, so the performance of the model for detecting both previously known and new attacks was demonstrated. The result obtained from the study show the effectiveness of the proposed hybrid approach that can be used to protect IIoT networks. Further developmental work will target optimality for real-time IIoT applications and enhancing the system’s applicability within vast IIoT systems.