Intelligent Intrusion Detection System for Drones Based on KNN and LDA Techniques

Abstract

Drones are being utilised in many domains such as disaster management, monitoring, emergency assistance, and military missions. Due to their missions, drones are exposing too many types of attacks such as Black hole, Grey hole, Wormhole, and other attacks. The security issue is a very important concern because drones are dealing with sensitive and expensive information, which should be protected and secured. In this thesis, a new intelligent intrusion detection system is designed and implemented to improve the security of the drone against the mentioned attacks for detecting malignant behaviors. The detection technique is presented for monitoring the drones' behaviors and distinguishing the normal and the malignant activities. However, K-nearest neighbors is utilised in this thesis which is a learning method based on instance type of objects classification depending on the training samples in the space of the features. The proposed Intrusion Detection System (IDS) passes through five stages which are: mobility generation stage, network simulation stage, data collection and preprocessing stage, training stage and testing stage. The dataset extracted from the trace file of the network simulator is used for testing the proposed system to measure its efficiency. The results show that the accuracy rate is 100% and the error rate is 0%. The low number of false alarms and the high accuracy rate prove that the proposed system is performing very well in detecting the attacks. The normalisation usage decreases the error rates while increases the detection rate. The detection is improved and the dataset problem is solved by performing the normalisation on the dataset that was produced from the trace file. The normalisation had a positive and direct effect on the experimental results by minimizing the false alarms and maximizing the accuracy detection rate. Linear discriminant analysis (LDA) technique is utilised in this work for minimizing dimensions. The results show that the accuracy rate is 99.71% and the error rate is 0.29%. Therefore, the proposed IDS is efficient and effective in distinguishing normal and malignant behaviors.