Near-Time Predictions of Future Truck Locations

Abstract

Location-based services are becoming an increasingly valuable service in the transportation sector. These location-based services rely on current location, but a better service can be provided by predicting the future location. So this thesis aims to investigate near-future location prediction of a specific vehicle using Recurrent Neural Networks (RNNs) and its different variants (a time-series approach). In addition, we discuss the required data pre-processing steps, the architecture of RNNs and the experiments to compare the different variants of RNNs. These experiments show that Long Short-Term Memory Networks (LSTM) have better predictions when compared to simple RNNs and Bi-directional Recurrent Neural Networks (BRNN). The results from this work will be helpful in developing a better short-term location prediction model and will provide better services to SCANIAs customers.