Correlation n-star Graphs

When working with correlation analysis and the visualization of its results and processes, the systematic identification of correlation classes is often overlooked. Such classes can be described as sets of attributes of the studied dataset that share similar correlation patterns − like the strength and direction of the relationship measured between a pair of attributes − and can be used as a method of identification of significant and insignificant correlations while introducing an attribute hierarchy for automated predictive analysis. This work proposes correlation n-star graphs, a graph-based visualization model designed to support automatic identification of these classes in multidimensional datasets. The work focuses on the design of the visualization model, a Python implementation of the proposed concept, and an experimental evaluation on three benchmark datasets assessing both qualitative and quantitative aspects of the visualization and correlation classification.