Indexing Method for Density-Based Clustering Accuracy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The DBSCAN algorithm is computationally expensive due to its O(n2) distance computations, making it inefficient for large data sets and limiting the exploration of parameter spaces for density-based clustering.

Innovation Solution

A computer-implemented method computes an index for density-based clustering, allowing for efficient exploration of parameter settings by selecting unprocessed records, computing core and reachability distances, and updating a priority queue to ensure exact clustering results.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If DBSCAN algorithm is applied to large data sets, then density-based clustering can be performed, but computation time becomes prohibitively long due to O(n2) distance computations

Engineering Contradiction:

Improveclustering accuracyVSAvoidcomputation time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent pre-computes and stores distance information in an index structure before actual clustering. By performing preliminary distance computations and organizing them in a sorted index, the algorithm avoids O(n2) computations during the clustering phase, reducing computation time while maintaining clustering accuracy

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent creates an index copy of the distance matrix that stores pre-computed distance information. This index structure serves as a simplified representation that can be queried efficiently without requiring access to the full distance matrix, thereby reducing computational overhead

Inventive Principle:

Principle #26Copying

2Productivity

If OPTICS algorithm is used to reduce computation time, then clustering can be performed faster, but exact clustering results are not guaranteed as border objects are wrongly identified as noise

Engineering Contradiction:

Improveclustering speedVSAvoidclustering precision

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The patent introduces an index structure as an intermediary between the data and the clustering algorithm. This index stores pre-computed distance information that enables the algorithm to efficiently determine cluster memberships and border objects without the approximations required by OPTICS, thus maintaining exactness while improving speed

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

By pre-computing and storing distance rankings in the index before clustering, the algorithm can efficiently evaluate border objects and their relationships to clusters without the approximations inherent in OPTICS. This preliminary organization of data enables both speed and precision

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250156447A1Method for creating an index for reporting large-scale variant clusterings

Publication Date: 2025.05.15 CELONIS SE

AI summary

Provided is a computer-implemented method for computing an index for a first density-based clustering of a collection of records, wherein the index is stored with a storage device. The index supports the extraction of exact clusterings for any selected threshold distance ε* less than or equal to a predefined threshold distance ε and a predefined number of records MinPts, which forms the pair of input parameters for which the index is computed.