Detection of Intrusive Attacks in Mobile Ad-hoc Networks based on ConvNext Model

Abstract

A Mobile Ad Hoc Network (MANET) is composed of multiple mobile nodes that operate without a fixed infrastructure, and the random mobility of nodes leads to changes in the network topology. However, the dynamic nature of MANET and intrusive attacks are major problems in MANET leading to loss the critical information. To overcome these limitations, a Deep Learning (DL) model ConvNext is proposed which is an improved version of the residual network incorporating transformer elements to detect intrusive attacks in MANET. The proposed ConvNext's deep architecture allows it to handle large-scale MANETs with complex attack patterns and adapt to different network topologies and traffic conditions, making it suitable for the dynamic nature of MANETs. Initially, the data are obtained by simulation in NS-3 and acquired from the KDD’99 cup dataset for comparison results. Then, these data are converted into numerical values by label encoding. Finally, the intrusive attacks in the MANET are detected and classified based on the attacks accurately. The experimental results of the proposed method achieved an accuracy of 99.5% for simulated data and 0.991 for the KDD’99 cup dataset which is higher than existing approaches such as Hybrid Adaboost-Random Forest (HARF) and Fuzzy Extreme Learning Machine (FELM).