<p>Internet of Things (IoT) is considered as a significant factor in communication networks and internet-based applications. Intelligent Transportation System (ITS) depends on IoT to link the sensors, devices, and other frameworks to provide real-world data investigation. The major drawbacks faced by conventional techniques are short transmission delays while transferring data among automobiles and Roadside Units (RSUs), highway security, and traffic congestion. Thus, it leads to delay, accidents, and huge traffic congestion as well as negative impact on the environment, thereby influencing the transportability of individuals. Traffic congestion due to the overutilization of automobiles is considered a major issue in urban regions. In order to solve these above mentioned issues, a new and adaptive deep learning model is proposed to manage the traffic congestion in IoT. Initially, the obtained input data is accumulated from the standard datasets. Consequently, the gathered data is fed into the model of Hybrid Adaptive and Attention Deep Learning Network (HAADLNet) to predict traffic data. The proposed network is performed by considering the Multi-Scale Capsule Network (MCapsNet) and Temporal Convolution Network (TCN). For maximizing the performance, the hyper-parameters in the proposed HAADLNet are optimized by Iteration-based Random Variable for Preschool Education Optimization Algorithm (IRV-PEOA). Therefore, the developed model efficacy is validated by distinct measures and validated with traditional approaches. Hence, the higher results attained can prove the better management of traffic congestion in IoT networks.</p>