A multiple kernel density clustering algorithm for incomplete datasets in bioinformatics.

Area of Science:

• Bioinformatics
• Data Science
• Machine Learning

Background:

• Bioinformatics datasets frequently contain missing attribute values, limiting the applicability of conventional clustering algorithms.
• Existing clustering methods often require complete data and manual parameter tuning, impacting accuracy and efficiency.
• There is a need for clustering algorithms that can handle incomplete data and automate parameter optimization.

Purpose of the Study:

• To propose a novel Multiple Kernel Density Clustering algorithm for Incomplete datasets (MKDCI).
• To address the challenges of missing data and parameter selection in bioinformatics clustering.
• To achieve a better trade-off between accuracy and efficiency in clustering incomplete datasets.

Main Methods:

• Developed the Multiple Kernel Density Clustering for Incomplete datasets (MKDCI) algorithm.
• Implemented missing value imputation for input data samples.
• Employed optimal kernel learning, dimensionality reduction, Isolation Forests for centroid detection, and arbitrary shape cluster assignment.

Main Results:

• The MKDCI algorithm effectively handles missing attribute values in bioinformatics datasets.
• It automatically learns an optimal kernel for clustering and reduces dimensionality.
• Demonstrated superior performance in producing high-quality clusters on incomplete data compared to existing methods.

Conclusions:

• The proposed MKDCI algorithm is effective for clustering incomplete bioinformatics datasets.
• MKDCI automatically optimizes parameters, leading to improved clustering quality without human intervention.
• It outperforms existing density and parameter-free clustering algorithms on incomplete data.