On Implementations of Bus Travel Time Prediction Utilizing Methods in Artificial Intelligence

Travel time prediction is an important part of intelligent transportation systems. This work is a continuation of a state-of-the-art review where the most prominent methods used for bus arrival prediction were investigated. An introduction where a brief overview of the process of data collection and a rationale for selecting the data sources is given. Subsequently, the process of setting up representative datasets is explained and a selection of AI methods is chosen. The Weka machine learning implementations are used to classify, and an analysis of k-Nearest Neighbor, Artificial Neural Networks, and Support Vector Regression is done with different parameters and attributes. The parameter analysis discovered the optimal parameters for the different classifiers on the datasets, while the attribute analysis investigated the use of weather, ticket, passenger and football match data to improve the prediction performance of the classifiers. An extensive group analysis investigated the performance of the classifiers on different training and testing periods. Finally, a proof of concept model for real-time prediction is presented and compared to an existing real-time system in Trondheim, Norway. The model is found to be competitive to the existing system.