Over the years, the behavioural modelling and the machine learning communities have progressed in parallel streams in the pursuit of predicting outcomes in mobility, health, environment, marketing and other domains. The interdisciplinary interest group will work on developing behavioural theory-driven machine learning approaches that combine elements of traditional behaviour modelling and machine learning in different domains. It will thus help in getting the best of both worlds.
Explaining the science
Historically, behaviour modellers have relied on mathematical models based on theories of psychology and economics to quantify how human decisions are affected by the attributes of the alternatives and characteristics of the decision-makers. These models are typically calibrated using small-scale survey data which are expensive to collect and prone to measurement errors and biases. On the other hand, the availability of large-scale passive data sources in recent years has led to wide-spread use of data-driven machine learning models which typically do not have a strong behavioural underpinning. This questions the applicability of the machine learning techniques in human behaviour prediction in the context of disruptive changes (e.g. radically new technologies, financial meltdown, etc.) where only the core behavioural principles are likely to hold, not the historic patterns of choices.
O1: Identify the strengths and weaknesses of machine learning and traditional behaviour models
O2: Contrast the methodologies used in machine learning and traditional behaviour modelling and formulate methods to combine them
O3: Apply the methodologies in different domains like mobility and urban analytics, health, environment, marketing and finance, etc.
What can the machine learning research community ‘learn’ from the behaviour modellers and vice versa?