Designing Decision Trees for Representing Sustainable Behaviours in Agents

Decisions made by workers in their daily routine have an environmental impact. The LOCAW project has analyzed the drivers and barriers for an employee to choose a particular option in large organizations. In this project, Agent-Based Models (ABM) seek to clarify interactions among relevant actors and provide insights into the necessary conditions to achieve more sustainable organizations. For theoretical and practical reasons, it was considered to use decision trees to represent the internal behavior of the agents in the model. This paper focuses on how to improve the generalization capabilities of these decision trees using feature selection and discretization techniques. The application of these techniques is intended to obtain simpler decision trees, but more accurate. Experimental results of three daily activities support the adequacy of the approach presented.