Optimization of UAV Intrusion Detection Based on LSTM-RNN

The increasing prevalence of Unmanned Aerial Vehicles (UAVs) has resulted in a rise in malicious activities targeting their networks, highlighting the critical need for effective Intrusion Detection Systems (IDS). This study introduces an innovative intrusion detection system for UAV communication that integrates Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) networks, complemented by various preprocessing algorithms to improve detection capabilities. To assess the system’s performance, a benchmark dataset was created by simulating real network traffic, and challenges related to dataset imbalance were tackled using techniques such as Synthetic Minority Over-sampling Technique (SMOTE), one-hot encoding, and min-max scaling. The evaluation on the CICIDS2017 dataset revealed that the model achieved a classification accuracy of $\mathbf{9 9. 8 4 \%}$, an F1-score of $\mathbf{9 9. 8 9 \%}$, a precision of $\mathbf{9 9. 9 9 \%}$, and a recall of 99.79%. This performance significantly surpassed that of Naive Bayes and other traditional methods, showcasing enhanced accuracy and reduced false positive rates.