A Novel Distributed Anomaly Intrusion Detection Model for Drone Swarm Network in Smart Nations

In the recent years, drones have been extensively used in variety of fields and given the essential nature of drone swarm services, such as network traffic monitoring and search and rescue operations, it is imperative to mitigate security vulnerabilities in the drone swarm network. Computational Intelligence in edge analytics has the ability to enhance predictive capabilities by expediting the conversion of high-level features into actionable insights for remote monitoring and triggering alarms during emergency incidents without depending on backend servers. This study represents a significant advancement in the development of intrusion detection techniques in drone swarm edge computing by proposing a distributed intrusion detection model in drone networks based on real-time data analytics framework utilizing hybrid deep LSTM-IG-SVM architecture. The architecture is validated on the CIDDS-2017 benchmark network attacks dataset. Initial layers of LSTM are employed to extract high-level features of sequential network packets. Information gain implemented on the top of LSTM is used for feature reduction accounting for the energy constraint and computing complexity of drones. The selected features are further used to train SVM for intrusion detection. It is concluded that the proposed hybrid architecture outperforms baseline LSTM with more than 99% of performance accuracy for intrusion detection. The problem of false alarming has also been resolved using the proposed architecture with false alarm rate observed to be as low as 0.1%.