Analyzing the Predictive Performance of Scheduler Pro- cess Models of Rule-based Activity-based Models

Abstract

Activity-based travel demand modeling systems to date can be classified into two modeling approaches Utility maximization-based econometric model systems, and Rule-based computational process model systems. This dissertation discusses three particular contributions with respect to rule-based computational process models specifically the ALBATROSS (A Learning-based Transportation Oriented Simulation System) models system. The first contribution is related to improving the predictive performance of the scheduling process models. The second contribution involves analysing performance boundaries for rule-based activity forecasting models. And the third contribution is to conduct a sensitivity analysis of the models at each decision step (decision tree models). To achieve the goals, the ALBAROSS model is integrated in the FEATHERS framework. FEATHERS (Forecasting Evolutionary Activity-Travel of Households and their Environmental RepercussionS) is an experimental framework developed to facilitate the development of modular activity-based models for transportation demand. To include ALBATROSS in FEATHERS (FEATALB), the model parameters were modified to fit the Flemish data. The ALBATROSS model and its components have been studied in details. However, some practical limitations were determined that restrained further experimentations and there was a need for new implementations. Some parts of the model were re-implemented. The implementation involved using technologies to boost the design of experiments conducted in this thesis. There are three major factors related to improve the predictive performance of rule-based models. The first factor is to ensure that the data are of good quality i.e. obtaining better data that are used to train the models at individual decision steps. The second factor involves utilizing better classifiers at individual decision steps that constitute the scheduling process model. The third factor is to achieve a better data representation in the context of reordering the decision steps in the process model.