Hybrid Model to Predict the Arrival Time at Tollgate of Vehicles on Expressway

Traffic congestion at toll booths is a pressing problem for service providers and highway users due to increased urbanization and car ownership. On expressways, the Electronic Toll Collection (ETC) significantly helps to reduce congestion. However, how regions/countries that do not have the conditions to apply ETC deal with this situation is still a big problem. The arrival time prediction of vehicles is one of the solutions. Researches on vehicles' arrival time on the expressways are minuscule, especially for heterogeneous vehicle flow. To the author's knowledge, they compare the effectiveness of traditional statistical models and modern machine learning methods. This trend leads to infrequent research focusing on combining models to take full advantage of them. Besides, the time to arrival of vehicles contains much volatility amongst different times of the day. Therefore, this study approaches the GARCH model to deal with periods of significant fluctuations in arrival time and proceeds to build a hybrid model of GARCH, LSTM. The hybrid model plays on the advantages of traditional statistical models and machine learning. The proposed model is analyzed on the E20 expressway, Japan's ETC2.0 traffic data set for one week transmitted by all types of vehicles. The results show that applying the proposed model can yield better predictive output.