An Energy‐Efficient and Smart Traffic Management Framework With Optimization and Deep Learning for VANET

ABSTRACT Vehicular Ad Hoc Networks, in short VANETs, a category of mobile ad hoc networks, facilitate Vehicle‐to‐Vehicle (V2V) and Vehicle‐to‐Infrastructure (V2I) communication for improved traffic management, safety, and autonomous driving. This work proposes a traffic flow prediction model that integrates metaheuristic optimization techniques such as Sand Cat Swarm Optimization (SCSO) and Raven Roosting Optimization (RRO) with the deep learning model Bi‐directional Long Short Term Memory with Stacked Autoencoder (Bi‐LSTM‐SA) to improve prediction accuracy. To optimize the Bi‐LSTM‐SA network, a hybrid SCSO‐RRO approach encodes neuron parameters like weights, biases, and activation functions. The SCSO explores the search space while RRO refines solutions by having ravens follow high‐fitness sand cats. A fitness function evaluates performance, and the process iterates until convergence, after which the optimized network is validated on a separate dataset. The proposed model Combined SCSO‐RRO‐Bi‐LSTM‐SA (C‐SC‐RRO‐Bi‐LSTM‐SA) is compared with existing algorithms such as Random Forest (RF), Artificial Neural Network (ANN), Bi‐LSTM‐SA, and evaluation parameters utilized to quantify the network's performance include accuracy, prediction, recall, F1‐score, and cross‐entropy loss.