Vehicular ad-hoc networks (VANETs) plays major role in intelligent transportation systems (ITS) that allowing vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication to enhance safety and traffic management. However, VANETs face challenges like dynamic topology, issues of high-traffic scalability, utilization of inefficient resource and communication delays. This study presents a comprehensive framework that merages Deep Learning (DL) models with simulation tools like SUMO and NS3 to face these challenges. One of the main elements of the work is the enhancement of Road-Side Units (RSUs) positions using a Gated Recurrent Units (GRUs) regression model. This model foretell the optimal deployment of RSUs via considering features such as coverage, transmission power, distance sensitivity and overlap. As a result, network performance is improved, proven via a Mean Squared Error (MSE) of 0.202 and Mean Absolute Error (MAE) of 0.154, along with enhancement in packet delivery ratios by 65.71%, decreased packet loss by 35.94% and maximized goodput by 141.33% and throughput by 137.76%. Also, the framework uses Convolutional Neural Networks (CNNs) for traffic classification through different scenarios, obtaining 100% accuracy, while GRUs are further used for regression assignments with a validation loss as low as 0.0011 for prediction of traffic flow and anomaly detection. Generally, the results present valuable perceptions into the interaction between traffic prediction, RSU placement and routing strategies, presenting a resilient, adaptive and scalable approach for improving VANETs and paving the way for efficient and more safer smart city transportation networks.