One of the major research domains of intelligent transportation systems (ITSs) is applying certain parameters, such as traffic flow volume and speed, to predict conditions on the freeway. However, the time series dataset cannot be directly applied to a machine learning algorithm, as the original time series dataset must first be transformed. The traditional transformation method is the sliding window, and a dataset transformed in this manner only generates the relation between continuous time periods. Therefore, we propose a weight-based data pre-processing mechanism for improving the Accuracy of the massive traffic flow prediction, in which the dataset is transformed by combining the historical and current continuous data. In this approach, the transformed dataset has a certain amount of weight between two kinds of data, with the nearest data having more weight than historical data. Using this method to improve traffic flow prediction accuracy, and applying three kinds of machine learning algorithms to validate the effects, the experimental results demonstrate that our method is superior to the traditional one.