Intelligent Detection System for FANETs using Machine Learning

Abstract

The lack of security is one of the primary problems in Flying Ad Hoc Network (FANET) that cannot be overlooked. With the network's rapid development and widespread use, the number of attacks trying to hack and damage drones is increasing. Therefore, finding practical solutions to this problem and securing the information provided by drones has become necessary to ensure human safety and maintain the confidentiality of its mission. Intrusion detection systems are more popular for detecting attacks and providing high security, but current systems lack high accuracy. Therefore, high false predictions open the way for new research through a security system to detect attacks with high accuracy. This thesis proposes an Intelligent Detection System (IDS) to detect spoofing and jamming attacks, one of the most lethal attacks on FANETs. The proposed security system implements Machine Learning (ML) algorithms to classify signals and uses the Unmanned Arial vehicles (UAV) attack dataset to implement and evaluate the proposed system. After obtaining the dataset, it is pre-processed to prepare for the classification phase. The extracted features from the dataset are inputs to the ML algorithms. In contrast, the output of ML algorithms is the classes in the label of the dataset. The ML algorithms used in this thesis are Decision Tree (DT), K_Nearest Neighbour (KNN), and Gradient Boosting (GB). After completion of the training, the proposed system is evaluated by calculating the confusion matrix and other performance metrics. Performance metrics show the DT algorithm's effectiveness, which provided 99.93% and 0.07% for vii accuracy, Error rate, and 100% for each f1_score, precision, and recall. In addition, DT achieved an execution time of 0.93 seconds, the lowest among all algorithms. Moreover, the proposed IDS is applied to a second dataset (CSE-CIC-IDS2018) to demonstrate its efficiency. In comparison, the DT algorithm provided an accuracy of 99.78% in classifying the signals, and this percentage is close to the classification results that IDS achieved with the UAV attack dataset.