Due to recent advancements in Internet of Things (IoT) and 5G technology, new and unprecedented opportunities have risen with the usage of Artificial Intelligence techniques (AI) for big data analytics. These developments further accelerate the digitization of human society in terms of smart cities and self-driving cars. In this cyberinfrastructure, the geospatial information has been proven to play a pivotal role in enabling researchers to develop novel theories and scientific knowledge in the domain of Intelligent Transport Systems (ITS). For the last couple of decades, researchers in the field of ITS have been trying to optimize the prediction of traffic data of road networks with variations in space and time. Since road networks have an inherent complexity due to their nonlinear characteristics, the prediction of traffic data remains a difficult task. Among the preceding works, the application of Convolutional Neural Networks (CNN) for prediction has been utilized to reflect the characteristics of traffic data on a road network with regards to time and space. However, previous studies were unable to incorporate the vanishing gradient as the depth of the model increased. This study aims to address this limitation with the use of a ResNet model constructed with transfer learning. Traffic data prediction for a road network with a complex structure was performed by combining the road link and Mobileye network sensor data. The ResNet models used in the study are ResNet-50 and ResNet-152. After fine-tuning, ResNet-152, the model achieved a higher improvement in accuracy by 3.8% compared with previous models with a relatively shallow structure. The model proposed in this study can be useful for real-time traffic data prediction of complex road networks. The current study performed traffic data prediction by considering only space and time. In future studies, it is necessary to include other factors that may affect the road network.