Calculation of estimated time of arrival using artificial intelligence

Abstract

Providing solutions for designing an optimal planning scheme is extremely important in transportation industry. A potential solution requires the development of models that are able to accurately predict the remaining travel time of trucks operating in a transport mission. Unfortunately this is not a trivial task as a transport mission is influenced by a variety of stochastic and impossible to predict factors. This study develops a variety of machine learning approaches and benchmark their ability to predict arrival times. In particular Support Vector Regression, Artificial Neural Networks, Gradient Boosting, Random Forest and Stacked Generalization models were developed for the aforesaid task. The proposed models are trained and evaluated using GPS and weather data for a transport mission between Malmö and Göteborg. The main objectives of the study are finding the variables that influence the total travel time of a vehicle and are optimal to be used as inputs to the prediction models, comparing the performance of different machine learning approaches and identifying the optimal approach among the proposed models. Study results verified that machine learning approaches have the ability to predict the arrival times of trucks. Even though all methods outperformed a historic data based model, results showed that the Random Forest and Stacked Generalization methods outperformed the other machine learning models in terms of Root Mean Square Error and Mean Absolute Percentage Error. In addition it was found that utilizing appropriate features as inputs to the prediction models dramatically increased the performance of the algorithms.