A Comparison Study on Similarity and Dissimilarity Measures in Clustering Continuous Data

Area of Science:

• Data Science
• Machine Learning
• Computational Statistics

Background:

• Similarity and distance measures are crucial for distance-based clustering algorithms.
• Existing research primarily focuses on low-dimensional spaces (2D-3D), leaving a gap in understanding performance in high-dimensional datasets.

Purpose of the Study:

• To propose a technical framework for analyzing, comparing, and benchmarking the influence of various similarity measures on distance-based clustering results.
• To address the lack of empirical studies on similarity measure performance in high-dimensional spaces.

Main Methods:

• Development of a technical framework to systematically evaluate similarity measures.
• Utilizing fifteen publicly available datasets, categorized into low and high-dimensional groups.
• Benchmarking the performance of different similarity measures within these categories.

Main Results:

• The study provides an empirical analysis of similarity measure behavior in high-dimensional clustering.
• Established a benchmark for evaluating existing and novel distance measures.
• Identified how different measures perform across low- and high-dimensional datasets.

Conclusions:

• The proposed framework facilitates the selection of appropriate distance measures for specific datasets.
• Enables reproducible comparison and evaluation of newly proposed similarity or distance measures against traditional ones.
• Contributes empirical evidence on the performance of similarity measures in high-dimensional data clustering.