A Combined Distance Metric Approach with Weight Adjustment For Improving Mixed Data Clustering Quality
Aung Pyae, Yeh-Ching Low, Hui Na Chua · 2024
Cluster analysis can be perceived as a problem of grouping data points according to their mutual similarity. Clustering quality largely depends on choosing an effective distance metric, especially when dealing with mixed data that contains both numerical and categorical features. This study introduces a Weighted Combined distance metric using Gower and sqrt-cosine distances to improve clustering quality for mixed data. By combining two individual distance metrics with weights, this unique technique generates an overall similarity score that can improve the clustering quality. Experiments have demonstrated that the Weighted Combined distance metric can yield superior clustering quality compared to individual distance metrics. These findings highlight the potential of this proposed method to effectively cluster mixed data and overcome challenges associated with mixed data clustering.