Honeycomb graphs for parametric identification of correlation classes in multidimensional datasets

In the process of gaining knowledge from large sets of data, one of the most significant methods from the area of descriptive statistics -correlation analysis -is applied to determine direct functional relationships between pairs of attributes.Even though the results of correlation analysis are measured through a crisp correlation coefficient, whose values belong to the [-1, 1] interval, human interpretation of these values is conventionally vague and uses linguistic classes of correlation to describe the strength of relationships between attribute pairs.However, this interpretative vagueness -and the correlation classes themselves -are not commonly employed in the decision-making processes.Therefore, this work focuses on the design and implementation of so-called Honeycomb Graphs -a visualization method for parametric identification of correlation classes in multidimensional datasets based on graphical models.After implementing the proposed visualization technique, two case studies on benchmark datasets are conducted, and the model is evaluated from both qualitative and quantitative points of view.The results of these studies highlight interactive exploration of correlation analysis while adhering to qualitative and quantitative standards of scientific visualizations and high utilization potential of the method in feature selection tasks, making it a valuable tool for predictive analysis and data exploration.