INTERNATIONAL CONGRESS ON RECENT ADVANCES IN SCIENCES AND TECHNOLOGY - Kuala Lumpur - Malaysia (2019-02-20)

Car Ownership Demand Modeling Using Machine Learning: Decision Trees and Neural Networks
Car ownership demand modeling is necessary for the travelling need analysis whether Trip or Tour was used as a unit of analysis. It exactly is the key factor to specify an individual’s or a family’s travelling behavior. Nowadays, there several two models to describe a car owner’s decision such as discrete choice model or machine learning model. Particularly, the machine learning model was designed to give a more certain prediction through different types of mechanism such as an Out-of-sample validation, a Non-linear structure, and an automated covariate selection. This article presents the car ownership demand model using 2 types of machine learning models including Decision Trees and Neural Networks with 3 answer classes: 0, 1, and 2+. The socio-demographic attribute with the impact on the household car ownership demand was also discussed. Prticularly, the machine learning model is able to select the variables automatically and assemble them as much as possible, so this article not only explores which of the socio-demographic factors is the key factor that perfectly reflects the household car ownership from both the decision trees and neural networks algorithms, but also compares the data sets to see the contrasts between those two models after adding the main attributes of the variables for both tour-based and activity-based models. The data used in this study was derived from the study on the suitability of the engineering, economics, finance, and environmental attributes with the impact under the project of the 2015 Khon Kaen expressway master plan for 2,015 households. The outcome indicated that the machine learning model could be used to predict the household car ownership as it was also found that when using the default parameters in all data sets, the neural networks algorithms provided the more correct result than did the decision tree algorithm. Nevertheless, in case of the household car ownership prediction from the data set with the added attributes of the key variable tested with both tour-based and activity-based models, the neural networks algorithm gave the similar result as found from the prediction using the data set with only the household socioeconomic status.
Mr. patiphan kaewwichian, Assoc. Prof. Ladda Tanwanichkul, Dr. Jumrus Pitaksringkarn