Travel Time Estimation Based on Neural Network with Auxiliary Loss

Yunchong Gan, Haoyu Zhang, Mingjie Wang

Abstract: Estimated Time of Arrival (ETA) plays an important role in various applications, for instance, scene of order dispatch, estimate price, travel time prediction, route decision, etc. In this project, we propose a new systematical Wide-Deep-Double-Recurrent model with Auxiliary loss (WDDRA), which involves Auxiliary Loss for Link Current Status prediction task. Our extensive evaluations show that WDDRA significantly outperforms the state-of-the-art learning algorithms. And our final ensemble model wins second place on the SIGSPATIAL 2021 GISCUP leaderboard without data augmentation. Our source code is available at:https://github.com/Phimos/SIGSPATIAL-2021-GISCUP-2nd-Place-Solution