A hybrid deep and machine learning model for short‐term traffic volume forecasting of adjacent intersections

Abstract

Despite complex fluctuations, missing data, and maintenance costs of detectors, traffic volume forecasting at intersections is still a challenge. Moreover, most existing forecasting methods consider an isolated intersection instead of multiple adjacent ones. By accurately forecasting the volume of short-term traffic, a low-cost method can be provided to solve the problems of congestion, delay, and breakdown of detectors in the road transport system. This paper outlines a novel hybrid method based on deep learning to estimate short-term traffic volume at three adjacent intersections. The gated recurrent unit (GRU) and long short-term memory (LSTM) bilayer network with wavelet transform (WL) noise reduction algorithm (WL+GRU-LSTM) are used to analyze raw traffic volume data. The WL+GRU-LSTM is constructed by comparing different machine learning and deep learning methods. A comparative study was used to choose the model’s network structure, training technique, and optimizer type. To prove the model’s accuracy and resilience, it was compared with the leading short-term traffic forecasting approaches. Experimental results confirm that the WL+GRU-LSTM model can forecast complex traffic volume fluctuations in different approaches of intersections with an accuracy of over 94%. It also shows better results compared to current methods. The proposed model could replace intermediate loop detectors.