Optimizing Traffic Prediction in Continuous Data Stream in Vehicular Networks with Triplet Transfer Learning and Dholes Hunting-Based Optimization

Traffic Congestion Detection Services (TCDSs) in network environments that use continuous data streams receive much information to detect and update road segments at low speeds and high vehicle density. Traditional traffic monitoring is laborious and inefficient. The biggest concern with manual monitoring is traffic controller security. Thus, in VANET (Vehicular Ad hoc Network), predicting traffic and monitoring congestion are essential to reduce delays and accidents. Numerous categorisation and prediction algorithms provide clear vehicle forecasts and collision-free network pathways. Traditional approaches have struggled to forecast path pictures accurately. Lack of prediction precision and sluggish processing speed prevent acceptable travel route decisions. We create a triplet transfer learning network (TTLN) to detect traffic congestion using input data to address these challenges. VGGNet will analyse input data for deep and statistical properties. A pre-trained triplet model represents three dimensions as an upstream transfer learning architecture task. Rebuilding the pre-trained model with the triplet model, temporal model, and auxiliary layer as the downstream job is the final step. Weights are fine-tuned to indicate traffic congestion. The suggested model’s classification accuracy is fine-tuned using Dholes Hunting-Based Optimisation (DHO). To evaluate our model against the most popular traffic prediction algorithms utilising categorisation criteria. The simulation results show that the suggested model exceeds the single method in time and prediction stability.