Learning and Identifying Human Behaviour Using Decision Trees
Stanislav Sitanskiy, Laura Sebastiá, Eva Onaindía · 2023
Recognition of activities in human-robot interaction typically assumes that humans follow a rational behaviour. However, how each person comes up with a plan depends on their own preferences, and usually, an optimal (rational) strategy is not always followed. This paper presents a method for learning human behaviours where the objective is to capture how the human selects the actions when solving a problem. To this end, we propose learning a Decision Tree that reflects a particular human behaviour using pairs state-action as training data where the state is represented by a set of automatically extracted features. A behaviour library is created with the learned decision trees, which are then used for identifying the behaviour followed by a person when executing a plan in a new situation. This identification is very useful for anticipating the person's needs and acting accordingly. Finally, we present an experimental evaluation with four different domains to test the validity of our approach.