This paper provides the most suitable method to improve urban traffic management by adapting traffic light timing based on actual traffic movements and vehicle density using regional convolutional neural networks (R-CNN). The current traffic light system relies on a set of predefined schedules, which are often not efficient and lead to increased traffic congestion. The proposed system leverages the power of R-CNN to accurately detect and count vehicles at intersections from a live CCTV camera, placed above the traffic light, enabling responsive adjustment of signal timings based on current traffic conditions. In addition to optimizing traffic lights, the system integrates data from probe vehicles equipped with GPS and communications systems to predict the status of upcoming traffic lights. By analyzing vehicle data, the system can predict signal changes and provide real-time recommendations to drivers. We introduced a machine learning technique focused on establishing a set of traffic light timing parameters using vehicle probe data. In this study, cycle estimation is carried out using the Light Gradient Boost model (LightGBM), while vehicle probe data is used to find red times per cycle using a neural network model. Red times per cycle will be calculated specifically for the car numbering phase from anonymous sources. These recommendations include whether a driver should wait at the current signal or adjust their speed to potentially bypass an upcoming red light by driving at an optimal speed, thereby minimizing stops and delays. Integrating R-CNN for density estimation and predictive analysis using probe vehicle data aims to improve traffic efficiency and reduce overall travel times. R-CNN is known for its high accuracy in object detection, especially in scenarios requiring precise localization and recognition. Thus, the demonstrated system has the potential to significantly improve urban mobility, reduce vehicle fuel consumption and emissions by reducing idle times at traffic lights.