CF Tree Data Clustering via Micro Cluster Segments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data clustering methods face challenges in efficiently clustering large datasets due to variability in sample representation and accuracy issues, particularly with sampling-based and dimension reduction-based methods, which can lead to inconsistent performance and missed important values.
Innovation Solution
A data clustering apparatus and method using a CF tree that performs effective clustering by generating micro cluster segments based on inter-microcluster distance and detecting directly or indirectly connected micro cluster segments through a range query, incorporating CF tree construction, micro cluster segment generation, and cluster set generation units to enhance clustering performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If sampling-based clustering method is used, then processing speed is improved, but clustering accuracy deteriorates due to poor sample representation
Solution Approach 1:
The patent segments the clustering process into multiple stages: first constructing a CF tree with micro-clusters from sampled data, then performing range queries to identify and merge similar micro-clusters. This segmentation allows efficient processing at each stage while maintaining overall clustering accuracy through systematic refinement.
Solution Approach 2:
The patent performs preliminary clustering by constructing a CF tree from sampled data before final clustering. The CF tree pre-organizes data into micro-clusters with computed centroids and radii, enabling efficient subsequent range queries and reducing the computational burden of final clustering operations.
2Device complexity
If dimension reduction-based clustering method is used, then processing complexity is reduced, but clustering accuracy deteriorates due to missed important values
Solution Approach 1:
The patent extracts only the essential features needed for clustering by computing centroids and radii for each micro-cluster in the CF tree. Instead of processing all original dimensions, the method works with these extracted summary statistics, reducing complexity while preserving clustering-relevant information.
Solution Approach 2:
The patent transforms the original high-dimensional data into a simplified representation using centroid coordinates and radius parameters for each micro-cluster. This parameter transformation reduces dimensionality while maintaining the essential spatial relationships needed for accurate clustering through range queries.
3Reliability
If traditional clustering method is used on large datasets, then comprehensive data coverage is improved, but processing time deteriorates
Solution Approach 1:
The patent uses sampling to process a partial set of data (sampled points) to construct the CF tree, rather than processing all data points. This partial action significantly reduces processing time while the range query mechanism ensures that the resulting clusters provide comprehensive coverage by identifying all regions with sufficient data density.
Solution Approach 2:
The patent replaces the traditional mechanical approach of processing all data points with a systematic algorithmic approach using CF trees and range queries. The systematic search through the CF tree structure efficiently identifies clusters without the computational overhead of examining every data point, substituting brute-force processing with intelligent algorithmic search.
Data Source
AI summary
Discloses are a data clustering apparatus and method based on a range query using a CF tree. The data clustering apparatus includes a CF tree construction unit configured to construct a CF tree, including a leaf node configured with a micro cluster (MC) and an index node configured with a sub-cluster (SC), based on a plurality of data included in a data set, a micro cluster segment (MCS) generation unit configured to generate an MCS based on a distance between the MCs while sequentially searching the leaf node to a right, and a cluster set generation unit configured to generate a cluster set by performing a range query on the MCS.