Internet of Drones Intrusion Detection Using Deep Learning

Abstract

Flying Ad Hoc Network (FANET) or drones’ technologies have gained much attraction in the last few years due to their critical applications. Therefore, various studies have been conducted on facilitating FANET applications in different fields. In fact, civil airspaces have gradually adopted FANET technology in their systems. However, FANET’s special roles made it complex to support emerging security threats, especially intrusion detection. This paper is a step forward towards the advances in FANET intrusion detection techniques. It investigates FANET intrusion detection threats by introducing a real-time data analytics framework based on deep learning. The framework consists of Recurrent Neural Networks (RNN) as a base. It also involves collecting data from the network and analyzing it using big data analytics for anomaly detection. The data collection is performed through an agent working inside each FANET. The agent is assumed to log the FANET real-time information. In addition, it involves a stream processing module that collects the drones’ communication information, including intrusion detection-related information. This information is fed into two RNN modules for data analysis, trained for this purpose. One of the RNN modules resides inside the FANET itself, and the second module resides at the base station. An extensive set of experiments were conducted based on various datasets to examine the efficiency of the proposed framework. The results showed that the proposed framework is superior to other recent approaches.

Smart Attacks Learning Machine Advisor System for Protecting Smart Cities from Smart Threats

“…Table 8 compares the SALMA and other state-of-the-art methods, using the KDD-CUP99 dataset [55]. Our proposed system was compared to different Machine Learning approaches [21,[24][25][26][27][28][29][30]. Our approach achieved the best accuracy.…”

“…In [26], DPC-GS-MND achieved the best results after SALMA, but it used all the features. In [27], RNN was used, but the techniques did not achieve high results as RNN is best suited for sequence models. Figure 12 compares the accuracy between the proposed system and other state-of-the-art methods.…”

“…They achieved 96.83% accuracy on the KDDCUP99 dataset. Ramadan et al [27] developed an IDS for Flying Ad Hoc Networks based on Recurrent Neural Networks (RNN), and achieved 91% on the KDDCUP99 dataset. Table 1 summarizes the related work, including the utilized technique, dataset, and the achieved accuracy.…”

“…They achieved 96.83% accuracy on the KD-DCUP99 dataset. Ramadan et al [27] developed an IDS for Flying Ad Hoc Networks based on Recurrent Neural Networks (RNN), and achieved 91% on the KDDCUP99 dataset.…”

An Aggregated Mutual Information Based Feature Selection with Machine Learning Methods for Enhancing IoT Botnet Attack Detection

“…Feature selection methods were used in different research disciplines to enhance the proposed machine learning models, for instance IDS for vehicular ad hoc networks [ 23 ], drone intrusion detection [ 24 ], clickbait detection on social media [ 25 ], detection of diseases in health informatics [ 26 ] and virtual screening for molecular similarity searching [ 27 ]. In addition to machine learning methods for IDS in IoT, several deep learning methods were applied for intrusion detection systems in IoT, which are discussed in [ 28 ].…”