Real-Time Kinematic (RTK) positioning is a precise positioning method that uses Global Navigation Satellite Systems (GNSS). It is applicable in challenging environments such as dense urban cities, especially in technologies like Intelligent Transportation Systems (ITS) or smart construction fields. However, since buildings, tunnels, and viaducts obstruct GNSS signals, RTK cannot provide consistent centimeter-level accurate solutions called “RTK Fix” in such locations. Moreover, it is difficult to predict the availability of the “RTK Fix” in an unfamiliar city. Therefore, we proposed a method to predict RTK availability based on a specific position and time, thus improving RTK user convenience. The proposed method uses a three-dimensional (3D) city model and machine learning. Based on satellite orbit information, the 3D city model was used to simulate the GNSS satellite environment at a specific location and time. These simulation results were used as explanatory variables when applying machine learning. The training data for machine learning were created by combining the measured RTK results for the city and the simulated explanatory variables. Furthermore, we conducted seven car test drives to obtain the measured RTK results for the urban area. We then separated these datasets into training and test datasets. Moreover, we used two machine learning algorithms: k-nearest neighbor (kNN) and logistic regression, to predict whether RTK is available. We evaluated the performances of the two proposed prediction algorithms via cross-validation and presented a Receiver Operating Characteristic (ROC) curve and the prediction accuracies of each dataset combination. The accuracy of the predicted RTK availability compared to the measured RTK availability was approximately 70–80%.