In the era of big data, traffic prediction, along with deep learning, has gained attention in previous years. Urban planners have long been concerned by the constant challenges of precisely forecasting traffic patterns. Conventional approaches, which mainly depends on statistical models, continually fails to accurately capture the dynamic and complex nature of vehicular traffic. These techniques encounters difficulties in fully encompassing the complex aspects that influence traffic dynamics. This study explores into the revolutionary potential of deep learning to address the limitations of existing methodologies, notably in addressing the intrinsic complexity of traffic forecast. We explore three different deep learning approaches - Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and a mix of both known as Hybrid. This review analyses their utility in decoding spatial and temporal complexities important for attaining accurate prediction. This research also emphasis on most frequently used datasets, acting as benchmark for assessing the robustness of traffic prediction models. This study seeks to impact the direction of future breakthroughs in applying deep learning to traffic prediction, adding to the ongoing refinement of its dynamic sphere. In the end we have highlighted some research challenges and future scope for the readers.