K-means Clustering Seed Initialization via Subsample Elimination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As data sets increase in size, number of dimensions, or number of clusters, the computational complexity of k-means clustering analysis increases, leading to longer computation times and increased resource demands, making it inefficient for timely data analysis in business applications.
Innovation Solution
The method involves preprocessing a sample set of data points to generate initial seeds for centroid calculations using subsample elimination techniques, reducing the number of iterations required for cluster centroid determination in multi-dimensional data sets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If k-means clustering analysis is performed on large data sets with many dimensions and clusters, then comprehensive data mining can be achieved, but computational complexity increases and computation time increases
Solution Approach 1:
The patent divides the complete data set into multiple subsamples, performs clustering analysis on each subsample independently to generate initial seed centroids, and then uses these seeds for the full data set analysis. This segmentation approach reduces the computational burden on any single operation while maintaining comprehensive coverage through the aggregation of results from multiple subsamples.
Solution Approach 2:
The patent performs preliminary clustering analysis on subsamples to pre-compute initial seed centroids before conducting the main clustering analysis on the complete data set. This preliminary action generates useful initialization values that accelerate convergence and reduce the computational complexity of the subsequent full data set analysis.
2Reliability
If k-means clustering analysis is performed on large data sets, then complete data patterns can be identified, but computation time increases
Solution Approach 1:
The patent segments the data set into multiple subsamples and performs clustering analysis on each subsample in parallel or sequentially to generate initial seeds. This segmentation enables the system to process data in smaller manageable chunks, reducing the time required for each operation while maintaining comprehensive pattern identification through the aggregation of results.
Solution Approach 2:
The patent performs preliminary subsample analysis to pre-compute initial centroids before the main analysis. This preliminary action significantly reduces the computation time required for the full data set analysis by providing good initial guesses that accelerate convergence, while still identifying complete data patterns through the comprehensive seed generation process.
3Reliability
If clustering analysis is performed on data sets with many dimensions, then comprehensive analysis is achieved, but computational complexity increases
Solution Approach 1:
The patent divides the high-dimensional data set into multiple subsamples, performing dimensionality reduction and clustering operations on each subsample independently. This segmentation approach manages computational complexity by processing lower-dimensional data chunks while maintaining comprehensive analysis through the aggregation of results from multiple subsamples.
4Ease of manufacture
If traditional k-means clustering is used without preprocessing, then simple implementation is maintained, but resource demands increase
Solution Approach 1:
The patent performs preliminary subsample analysis to pre-compute initial seed centroids before the main clustering analysis. This preliminary action reduces the computational resources required during the main analysis by providing good initial guesses that accelerate convergence, thereby reducing overall energy consumption while maintaining implementation simplicity through a straightforward two-stage process.
Data Source
AI summary
Improved efficiencies of data mining clustering techniques are provided by preprocessing a sample set of data points taken from a complete data set to provide seeds for centroid calculations of the complete data set. Such seeds are generated by selecting a uniform sample set of data points from a set of multi-dimensional data and then seed values for the cluster determination calculation are determined using a centroid analysis on the sample set of data points. The number of seeds calculated corresponds to a number of data clusters expected in the set of multi-dimensional data points. Seed values are determined using subsample elimination techniques.