A New Distance Metric Exploiting Heterogeneous Interattribute Relationship for Ordinal-and-Nominal-Attribute Data...

Area of Science:

• Data Science
• Machine Learning
• Statistics

Background:

• Clustering mixed-categorical data presents challenges due to existing distance metrics ignoring ordinal information or assuming attribute independence.
• These limitations can distort attribute distances and lead to suboptimal clustering outcomes.

Purpose of the Study:

• To develop a new entropy-based distance metric for mixed-categorical data clustering.
• To address the drawbacks of existing methods by preserving ordinal relationships and utilizing attribute correlations.

Main Methods:

• An entropy-based distance metric is proposed that quantifies category distances by exploiting correlations between attributes.
• The metric preserves the inherent order of ordinal categories and weights attribute contributions based on their interdependence.
• It unifies distance measurements for both ordinal and nominal attributes.

Main Results:

• The proposed metric effectively overcomes the limitations of existing methods in mixed-categorical data clustering.
• It successfully preserves ordinal information and incorporates attribute correlations into distance calculations.
• The metric is parameter-free and does not increase computational cost.

Conclusions:

• The novel entropy-based distance metric offers a superior approach for clustering mixed-categorical data.
• It provides a unified and accurate way to measure distances, leading to improved clustering performance.
• The method is computationally efficient and easy to implement.