Developing a disaggregate travel demand system of models using data mining techniques

The travel demand modelling has experienced a paradigm shift from aggregate to disaggregate models, leading to an increase in computational time and simulation cost. Meanwhile, transferability models have emerged to reduce the associated cost and computational burden, but haven't discounted the disaggregation level. This research proposes the proof of the concept of an innovative transferability modelling framework to estimate total number of trips and trip attributes in a tour of trips at a disaggregate level. In contrast to tour-based or activity-based models, the focus of transferability models is on replicating trip patterns rather than reflecting travellers' behaviour. Similar to previous transferability models, classifying decision tree is utilized as one of the modelling techniques in this study. Moreover, the merits of a modified version of decision tree and the random forest methods are examined. Victorian Integrated Survey of Travel and Activity (VISTA) in 2007 and 2009 are utilized to calibrate and validate the proposed framework, respectively. According to the results, the random forest method shows highest individual-level accuracy while matching the system-level observed distributions.