Enhancing IoT intrusion detection using an LSTM-based deep learning model with GAN and explainable AI

The enormous growth of IoT networks and security risks has encouraged research into intrusion detection, which is based on deep learning methods. IoT network security is an essential concern, and intrusion detection solutions are required to secure the network. In this study, we developed an LSTM-based model to detect various attacks in the IoT system. The filter correlation approach is used for selecting features, while data augmentation techniques SMOTE and GAN are used to address the class imbalance issue. The Pearson correlation method reduces the features after selecting one feature out of two having a correlation value more than the 0.95 (threshold) value. The minority attack classes DoS and Fuzzers are increased by generating synthetic data using data augmentation techniques SMOTE and GAN. The model used the UNSW-NB 15 dataset and is evaluated on various parameters. The findings indicated that the GAN-LSTM model attained an accuracy of 92%, outperforming other models with high precision values, and it improved both the overall accuracy and the accuracy of the minority class. The explainable AI methods LIME and SHAP are implemented to explain the model prediction and also find the prominent feature to build trust in the model prediction.