A New-Fangled FES-k-Means Clustering Algorithm for Disease Discovery and Visual Analytics

Area of Science:

• Data Science
• Computational Biology
• Geospatial Analysis

Background:

• The original k-means clustering technique can be computationally intensive for large datasets.
• Efficient data clustering is crucial for knowledge discovery and data mining applications.
• Geospatial data analysis has significant implications for understanding disease patterns and mechanisms.

Purpose of the Study:

• To evaluate the performance and quality of a new algorithm, the Fast, Efficient, and Scalable k-means (FES-k-means) algorithm.
• To provide evidence for FES-k-means as an improvement over the original k-means clustering technique.
• To explore the application of FES-k-means in analyzing large geospatial datasets for disease mechanism discovery.

Main Methods:

• Developed the FES-k-means algorithm, a hybrid approach combining k-d tree data structure, the original k-means algorithm, and Mashor's adaptation rate.
• Tested the algorithm on two real-world datasets and one synthetic dataset.
• Applied the algorithm in a two-step process: first on data trained by the MIL-SOM method, then on untrained data.

Main Results:

• The FES-k-means algorithm produced clusters comparable to the original k-means method.
• Runtime comparisons demonstrated significantly faster performance for FES-k-means.
• The algorithm efficiently analyzed large geospatial datasets, showing potential for disease mechanism discovery.

Conclusions:

• FES-k-means offers a substantial improvement in speed and efficiency for k-means clustering.
• The two-step training and clustering approach enhances knowledge discovery capabilities.
• The algorithm's efficiency in geospatial analysis may aid in discovering links between environmental exposures, such as water service lines, and disease patterns, like elevated blood lead levels in Chicago.