Optimizing Decision Trees for Enhanced Human Comprehension

This paper studies a novel approach for training people to perform complex classification tasks using decision trees. The main objective of this study is to identify the most effective subset of rules for instructing users on how to excel in classification tasks themselves. The paper addresses the challenge of striking a balance between maximizing knowledge by incorporating numerous rules and the need to limit rules to prevent cognitive overload. To investigate this matter, a series of experiments were conducted, training users using decision trees to identify cases where cancer is suspected, and further testing is required. Notably, the study revealed a correlation between the decision tree characteristics and users’ comprehension levels. Building on these experimental outcomes, a machine learning model was developed to predict users’ comprehension levels based on different decision trees, thereby facilitating the selection of the most appropriate tree. To further assess the machine learning model’s performance, additional experiments were carried out using an alternative dataset focused on Crohn’s disease. The results demonstrated a significant enhancement in user understanding and classification performance. These findings emphasize the potential to improve human understanding and decision rule explainability by effectively modeling users’ comprehension.