Intelligent Transportation Systems (ITS) has been developed to aid drivers and other road-users to make a better travel decision. In recent years, many research efforts have been devoted in this field. Being one kind of time-series data, we can analyze the traffic data following the general aspects of studying time-series, which contains the analysis of periodicity of many kinds. This work highlights the study on the (long-term) multiple periodicities that could be found in traffic data while also considers more specific aspects such as unexpected short-term patterns, spatial relationship and feature correlations. Thanks to the periodicity of traffic data, most experienced drivers can tell how the traffic state will be on the road with given specific time and location. We aim to propose an approach with many of the above aspects to reach a quality traffic speed forecasting. We choose Gaussian process regression as the base model to realize the approach. Given the forecasting that considers all the above aspects, we enjoy the speed forecasting performance with MAE equal to one to two mph at its peak performance for a challenging speed forecasting 30-minute ahead of the current time.