Intrusion Detection in IoT Networks Using LSTM Deep Learning Models with the UNSW-NB15 Dataset

Network Intrusion Detection Systems (IDS) are an essential part of the Internet of Things (IoT) Networks, addressing the increasingly severe challenges posed by Internet security threats. This research enhances (IDS) capabilities in IoT environments by investigating the performance of Deep Learning Models with Binary Classification for identify and predict if the connection is normal or attack. The Deep Learning Models; including Long Short-Term Memory (LSTM), Bidirectional LSTM, Gated Recurrent Unit (GRU) and Bidirectional GRU. The comprehensive UNSW-NB15 Dataset is utilized that includes real-world network traffic and a variety of attacks. To effectively tackle the issue of Class Imbalance in the Dataset, we utilize the Synthetic Minority Over-sampling Technique (SMOTE). This approach enhances the representation of minority classes, leading enhances our model's performance and robustness. We train and evaluate these models, our evaluation based on Accuracy, Recall, Precision, F1-score and AUC metrics. The results indicate that the model with the best metrics is the Bidirectional LSTM BiLSTM, achieving an Accuracy of 0.9468, AUC of 0.9801 without SMOTE and an Accuracy of 0.9501, AUC of 0.9953 with SMOTE. These findings emphasize the effectiveness of the BiLSTM model for a more accurate and robust Intrusion Detection Systems (IDS) using the SMOTE technique and underscore the beneficial impact of the SMOTE technique on class imbalance issues. The research contributes valuable insights towards the development of more robust and secure IoT systems using the LSTM model using the SMOTE technique.