DDCAL: Evenly Distributing Data into Low Variance Clusters Based on Iterative Feature Scaling.

Area of Science:

• Data Science
• Computer Science
• Statistics

Background:

• Clustering one-dimensional data for visualization presents challenges, particularly in handling outliers and detecting small clusters.
• Existing methods may overemphasize outliers, hindering effective data representation on choropleth maps and process models.

Purpose of the Study:

• To develop and evaluate a novel heuristic algorithm for clustering one-dimensional data with an emphasis on even distribution and low variance.
• To improve data visualization techniques by minimizing outlier impact and enhancing the detection of smaller data clusters.

Main Methods:

• Introduction of the 1d distribution cluster algorithm (DDCAL), a heuristic approach utilizing iterative feature scaling.
• Testing DDCAL on 5 artificial and 4 real-world datasets with diverse distributions and use cases.

Main Results:

• DDCAL demonstrates stable clustering results, generating evenly distributed clusters with low variance.
• Comparative analysis shows DDCAL outperforming 11 existing clustering algorithms on tested datasets.
• Successful application in visualizing pandemic and population data on choropleth maps and process mining results.

Conclusions:

• DDCAL offers a robust solution for one-dimensional data clustering, particularly beneficial for data visualization applications.
• The algorithm's ability to handle outliers and detect small clusters enhances the interpretability of complex datasets.