Improving intrusion detection using LSTM-RNN to protect drones’ networks

Abstract

The expanding use of Unmanned Aerial Vehicle (UAVs)/drones has been noticeable in recent years. Drones have several uses in a wide range of industries, including the military, delivery, agricultural, and surveillance. This led to a visible increase in malicious activities targeting drones’ network. Consequently, it has become imperative to develop intrusion detection systems. The network intrusion detection system (NIDS) uses deep learning to identify network anomalies. In this paper, a new approach is proposed to enhance IDS in drone communications. The proposed model utilizes the Recurrent Neural Network (RNN) with a Long Short-Term Memory Network (LSTM) combined with pre-processing algorithms. Simulating real network traffic was necessary to do benchmark datasets to evaluate the IDS performance. Due to the artificial part in datasets, there is unbalancing between the normal and attack traffic. Training models on high-dimensional datasets with redundant features can be computationally expensive, need more storage, and lead to low performance. The cleaning of the dataset is accompanied by the most effective pre-processing techniques. SMOTE for unbalancing, one-hot encoding, and min–max scaling techniques are used to mitigate the dataset issues. The model is evaluated using the most up-to-date version of the dataset CICIDS2017 (13 May 2023). The model successfully achieves 99.84 % classification accuracy, 99.84 % F1-score, 99.99 % Precision, and 99.70 % recall. The proposed model outperformed the Naïve Bayes and five other legacy protocols in accuracy and False Positive rate.

Keywords

• Unmanned Aerial Vehicles (UAVs)
• Network Intrusion Detection System (NIDS)
• Recurrent Neural Network (RNN)
• Deep learning
• Long Short-Term Memory (LSTM)
• Naïve Bayes