Computational concept for human food choice and eating behaviour

Abstract

An inadequate diet can cause a number of illnesses with some of them posing major threats for humanity. Poor diet largely originates from behavioral and social issues rather than environmental factors. With simulation being a grand tool to analyze and address behavior issues, relatively few studies focus on computational modeling of nutrition at behavioural level. Furthermore, we have overviewed several popular approaches to computational modeling and simulating dietary decision-making and found no clear favorite. Further still, modelers rarely pay attention to one of the key behavioural factors – motivation. In the vast majority of models, either motivation is assumed to be exogenously given and, hence, is left out of the model, or motivation is not taken into account in any form, even though ignoring incentives significantly reduces adaptive capabilities of any human-like goal-directed model entity. We aimed to outline a modelling approach that would fit into the food choice topic and would improve on the available models. This implies offering an intelligible algorithm that would be easily applied to statistical data yet offering a depth of analysis despite its seeming simplicity. Thus, we present our view of the food choice simulation problem which employs eating incentives and an original choice mechanism that is different both from traditional maximizing approaches common to economics and artificial intelligence research and from the dominant psychological computational approaches. We outlined the programming conceptual algorithm that involves sequential incentive (which can result from the biological necessities, social, intellectual or spiritual needs alike) selection, incentive-foodstuff coupling (a relation can be either fixed or dynamic) and elimination of undesirable food options based on incentives ranking (qualitative ranking seems to be preferable over quantitative ranking, forasmuch as it resembles the way of thinking of a regular person more closely) supplemented by pseudocode segments. The algorithm suits agent-based simulation paradigm, yet it is not tied to it and can be fitted with other simulation approaches as well. The algorithm is supposed to be implemented in Java. Since the offered algorithm is conceptual it requires an implementation to bring about robust conclusions which is our goal to reach next.