An Analysis of Human Perception of Partitions of Numerical Factor Domains

In Machine learning (ML), several discretization techniques and mathematical approaches are used to partition numerical data attributes. However, cut-points retrieved by discretizing techniques often do not match with human perceived cut-points. Therefore, understanding the human perception for discretizing the numerical attribute is important for developing an effective discretizing technique. In this paper, we conduct a study of human perception of partitions in numerical data that reflects best the impact of one independent numerical attribute on another dependent numerical attribute. We aim to understand how expert data scientists and statisticians partition numerical attributes under different types of data points, such as dense data points, outliers, and uneven random points. The findings lead to an interesting discussion about the importance of human perception under distinct kinds of data points for finding partitions of numerical attributes.