CLAMS: A Cluster Ambiguity Measure for Estimating Perceptual Variability in Visual Clustering

Area of Science:

• Data Visualization
• Human-Computer Interaction
• Perceptual Science

Background:

• Visual clustering in scatterplots is crucial for data analysis but suffers from individual perceptual variability.
• Ambiguous cluster boundaries and individual differences lead to unreliable visual clustering results.
• Existing methods lack systematic approaches to assess perceptual variability in visual clustering.

Purpose of the Study:

• To systematically study and quantify perceptual variability in visual clustering, termed Cluster Ambiguity.
• To introduce CLAMS, a data-driven visual quality measure for predicting cluster ambiguity in monochrome scatterplots.
• To provide a reliable method for assessing the ambiguity of visual clusters.

Main Methods:

• Conducted a qualitative study to identify factors influencing visual cluster separation (e.g., proximity, size).
• Developed a regression module to estimate human-judged separability between cluster pairs.
• Aggregated pairwise separability scores to predict overall cluster ambiguity using CLAMS.

Main Results:

• CLAMS accurately predicts ground truth cluster ambiguity, outperforming existing clustering techniques.
• CLAMS demonstrates performance comparable to human annotators in assessing cluster ambiguity.
• The measure effectively quantifies perceptual variability in visual clustering tasks.

Conclusions:

• CLAMS offers a reliable and efficient method for assessing cluster ambiguity in scatterplots.
• The developed measure can be used to optimize and benchmark data mining techniques.
• This work enhances the trustworthiness of data analysis relying on visual clustering.