Accurate prediction of road traffic speed has a crucial impact in estimating traffic condition and plays a role in optimizing transportation and traffic system’s function. However, the nonlinear nature of traffic systems and the complexity of uncertainty introduce challenges for speed variables. Therefore, finding hidden patterns in traffic is the most critical issue in predicting traffic speed. This research aims to predict the traffic flow in the street highways using a new a hybrid model. Previous methods proposed to address these challenges are fundamentally limited in providing optimal solutions due to their inability to capture both local and global nonlinear patterns accurately. To overcome these limitations, this paper proposes a method that combines Convolutional Neural Networks (CNNs), Temporal Convolutional Networks (TCNs), and Long Short-Term Memory (LSTM) networks, denoted as CNN-TCN-LSTM. The incorporation of CNNs aims to effectively extract localized features within the data. Concurrently, Recurrent Convolutional Networks (RCNs) and Long Short-Term Memory (LSTM) networks are employed to model both local and global temporal dynamic dependencies. The dataset comprises information obtained from loop inductive detectors deployed along the freeways within the Seattle metropolitan region during the year 2015. Data was gathered from a total of 323 sensor stations positioned along the designated route. The evaluation of the proposed model shows a performance and accuracy improvement of 1% and 9% compared to LSTM and RNN-based prediction methods, respectively.