Asymmetric binary similarity measures

Area of Science:

• Ecology
• Data Science
• Bioinformatics

Background:

• Asymmetric binary data, where one state is more informative (e.g., presence vs. absence), poses challenges in classification.
• Group homogeneity, crucial for accurate classification, is defined by shared informative states and minimized mismatches.
• Existing similarity measures may not optimally handle data asymmetry, impacting classification accuracy.

Purpose of the Study:

• To evaluate the behavior of common binary similarity measures concerning 1-state homogeneity.
• To introduce and validate a novel coefficient,
• C
• designed to overcome limitations of existing measures.
• To provide recommendations for using binary similarity coefficients in ecological and other classification contexts.

Main Methods:

• Analysis of common binary similarity coefficients, including the Jaccard coefficient.
• Introduction and theoretical examination of a new coefficient,
• C
• .
• Evaluation of coefficient performance in relation to 1-state homogeneity and group classification.

Main Results:

• The Jaccard coefficient approximates desired behavior but shows limitations with certain data values and match/mismatch ratios.
• The new coefficient,
• C
• effectively addresses the identified issues with asymmetric binary data.
• Coefficient
• C
• demonstrates superior performance in achieving homogeneous classifications.

Conclusions:

• Existing binary similarity measures have limitations when dealing with asymmetric data crucial for ecological classification.
• The newly proposed coefficient,
• C
• offers a more robust and accurate approach to classifying individuals based on asymmetric binary characters.
• Recommendations are provided for the optimal application of binary similarity coefficients, emphasizing the utility of coefficient
• C
• .