Learning User Preferences of Route Choice Using Fuzzy Decision Tree Induction
Kyounga Park, Michael G H Bell · Transportation Research Board 90th Annual MeetingTransportation Research Board · 2011
Decision tree induction, one of machine learning techniques, can be used to build a route choice model by finding patterns observed in repeated route choice behaviour. Although a decision tree successfully accommodates route choice preferences, the use of “sharp cut-off points” to discretise the domain of a continuous-valued attribute makes a decision tree too sensitive, leading to misclassifications. This paper introduces fuzzy decision tree induction to relax the sensitivity of classical decision trees. Transforming crisp discretisation in a classical decision tree into soft discretisation using fuzzy representation, fuzzy decision tree induction increases flexibility in classifying a new instance. The experiment results indicate that fuzzy route choice decision trees outperform non-fuzzy decision trees in terms of predictive accuracy and that the fuzzy models are more effectively applicable in practice.