K Nearest Neighbor and Flexible Neural Tree Based IDS in Mobile Ad- hoc Network

Abstract

Mobile ad-hoc networks (MANETs) have brought about a lot of importance to recent researches because of its popularity and developing benefit. However, they seem to be defenseless against the various security threats that decrease their efficiency in comparison to other networks, by all accounts. Intrusion Detection Systems (IDS) creates a second line of safeguard against the various vulnerabilities to MANETs, as they scan the network for malicious activities performed by attackers. Because of the distributed characteristic of MANET, traditional cryptography mechanisms cannot shield MANETs entirely as far as novel attacks and vulnerabilities are concerned. Hence by applying machine learning strategies for IDS, these difficulties can be resolved. In this paper, numerous research works based on ML techniques over four types of attacks (DOS, Probe, U2R and R2L) have been reviewed and their performance is observed. From the above reviewed work, it can be inferred that KNN and FNT algorithm would perform better than the rest. After applying KNN and FNT in the proposed intrusion detection MANET model, KNN performed better than FNT. Accuracy of KNN for detection of DOS, Probe, U2R and R2L was 99.24%, 99.13, 98.89% and 98.42% respectively, whereas accuracy of FNT for detection of the same attacks was 98.35%, 98.07%, 97.84% and 98.01% respectively. So the detection accuracy of KNN is better than FNT. For KNN, the TP, FP, FN and precision value is 0.931, 0.015, 0.063 and 0.983 respectively , whereas for FNT, the TP, FP, FN and precision value is 0.815, 0.153, 0.274 and 0.826 respectively. From these above results, it is clear that KNN is better than FNT as the True Positive value and Precision is higher while the False Positive value and False Negative value is lower for KNN.