Machine Learning Driven Threat Identification to Enhance FANET Security using Genetic Algorithm

Abstract

Emerging as a highly promising technology, Flying Ad-hoc Networks represent self-organizing networks of Unmanned Aerial Vehicles (UAVs), garnering attention for their diverse applications spanning environmental monitoring, disaster management, precision agriculture, surveillance, and military operations. However, these networks face challenges to various security threats, including malicious node detection due to their deployment in dynamic environments. To address this issue, we present an improved novel security solution, Machine Learning-based Threat Identification for FANET using a Genetic Algorithm (ML-TIFGA) in this paper. The research includes the detection of abnormal behavior nodes using a basic genetic algorithm and dynamically adapting the changing network conditions by utilizing a reputation system. To enhance our security solution ML-TIFGA, we evaluated two key factors: cooperation and trustworthiness, which act as genetic elements within the chromosome of the flying node in our genetic population. Further, a mechanism is incorporated to reconfigure the trust, addressing the challenge of dynamically extracting threats through the updated weighted reputation system while considering past behavior monitoring. Significant improvements were found in the experimental results using actual sample values from the NSL-KDD dataset, which produced a remarkable 99.829% classification accuracy. Additionally, threat identification rates reached 98.36% for training and 98.86% for testing samples, with a remarkable improvement of 99.3% in network reliability through ML-TIFGA. When benchmarked against state-of-the-art approaches, performance metrics such as delay, throughput, and data delivery rate exhibited notable enhancements of 24.65%, 29.16%, and 31.73%, respectively.