Adaptive data processing framework for efficient short-term traffic flow prediction

Accurate short-term traffic forecasting is a prerequisite for establishing intelligent transportation systems. In this paper, a new adaptive traffic flow sequence framework is processed. For any given dataset, the framework divides the original sequence into different time periods, generating an adaptive traffic flow sequence. Compared with the sequence using fixed aggregation intervals, adaptive traffic flow data is reduced by 33% on average. Subsequently, local and global prediction models are presented for adaptive traffic flow sequences. Namely, the data-extended GRU algorithm and data-adaptive Bi-GRU algorithm respectively demonstrate best mean absolute percentage error of 4.70% and 6.80% according to the data experiments.