Generalization analysis of ANN-Based routing protocols for diverse vehicular environments in VANETs

Abstract

The rapid advancement of intelligent transportation systems depends on effective and secure routing within vehicular communication networks under diverse driving conditions. This research investigates the efficacy of neural network-based routing protocols in enabling reliable and secure data transmission under diverse traffic conditions. Prior research has shown that neural networks can alleviate security threats, such as malicious node attacks; however, there has been inadequate exploration of their adaptability in urban, rural, and highway environments. This research examines routing performance in simulated traffic environments and actual mobility data to address this gap. The methodology utilizes robustness testing, transfer learning, and cross-domain validation to evaluate the sensitivity of routing models to variations in vehicle density, mobility patterns, and road configurations. The findings indicate that the neural network-based approach outperforms a conventional routing protocol across various contexts. In urban areas, the delivery rate increased from 78 % to 85 %, while in rural regions, it rose from 65 % to 77 %. We reduced the end-to-end delay by approximately 7 to 12 milliseconds in all instances. Relative to the baseline, throughput increased by approximately 10 to 15 percent, while energy efficiency improved by 5 to 8 percent. The proposed method enhanced system resilience against attacks, successfully thwarting over 90 % of malicious disruptions, in contrast to the 73 to 79 % efficacy of the previous protocol. This study presents a framework for designing adaptive and scalable routing systems that maintain consistent performance across diverse vehicular conditions. The findings enhance the safety and efficacy of intelligent transportation systems.