Compact Class-Conditional Attribute Category Clustering: Amino Acid Grouping for Enhanced HIV-1 Protease Cleavage...

Area of Science:

• Bioinformatics
• Machine Learning
• Data Science

Background:

• Categorical attributes pose challenges in classification tasks as category numbers increase.
• High cardinality attributes negatively impact model building time, complexity, and performance.

Purpose of the Study:

• To propose a novel preprocessing technique for grouping attribute categories in classification datasets.
• To mitigate issues associated with high cardinality categorical attributes.

Main Methods:

• Combines Euclidean space representation of categorical associations, clustering, and attribute quality metrics.
• Groups similar attribute categories based on their contribution to classification.
• Evaluated on HIV-1 protease cleavage site prediction using amino acid attributes.

Main Results:

• Achieved significant reduction in categories per attribute (74%-81%) on HIV-1 datasets.
• Observed improvements in classification performance: up to 0.07 in accuracy and 0.19 in geometric mean.
• Validated robustness through extensive simulations on synthetic datasets.

Conclusions:

• The developed method effectively simplifies data representations and enhances classification performance.
• Demonstrates capability to improve HIV-1 cleavage prediction, aiding viral process understanding and therapeutic strategies.