Activity type inference and universal patterns for activity sequence generation in lightweight transportation models

Abstract

Although transportation-related challenges have existed throughout the ages, more recently they have assumed enormous proportions. Negative economic, social and ecologic consequences are the result of our socio-economic system seeking a continuous economic growth through an increase in economic activities (e.g. work-related personal transport, logistics), and our desire for comfort while moving from one place to another. Technological solutions are not always straightforward or possible at all, or are eventually not sustainable because of rebound effects. Travel behavior (as the outcome of i.a. our personal desires and obligations towards third parties), infrastructure and technology are strongly linked to the transportation-related challenges. Cities are growing, escalating the need for novel and thoughtful decision making at policy level for these concentrated living areas. Infrastructural and digital innovations are introduced into our lives at an increasing rate, making tools to understand their (long-term) consequences a necessity. Transportation demand models are such a tool. They provide information on the mobility and activities of people, and the effects that policies or infrastructural changes have on them. They, however, have some challenges of their own. They are complex and require a lot of data in order to understand people's travel behavior and the effects of infrastructure. This dissertation attempts to provide extra tools and knowledge to make these models more readily available to city councils, governments and academia. Its first aim is to make the supply of activity-travel data, being the input data for transportation models, more easy. It does this by introducing a model to estimate travel purpose from context-lacking GPS data, which may be available in large quantities from already omnipresent mobile devices. The predictive power of using such a tool is optimized as well. The second aim is to make it easier to generate activity sequences, the patterns of our daily rhythm. They are the source of our need for transportation, and thus an important component in advanced transportation models. A first proposed technique finds `skeletal activity sequences', which may aid the activity sequence component in transportation models. A second technique aspires to contribute to data-less transportation models, which are based on universal mobility patterns. Universal patterns related to the formation of activity sequences were demonstrated in an international scope, and resulted in a novel activity sequence generation framework based on universal patterns, therefore requiring hardly any data.