Density-Based Clustering Algorithm for Multidimensional Data

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing clustering algorithms for large datasets, such as those in biological and chemical sciences, face limitations including inability to handle varying densities, inconsistency due to input order, and inefficiency, particularly in forming clusters without iterative processes.

Innovation Solution

A density-based clustering method with a single adjustable parameter (cutoff distance) that identifies clusters without forcing a specific shape or number of clusters, is noise-tolerant, and executes faster by determining cluster centers through local density maxima, allowing for practical computation on large datasets.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If iterative procedures are used to determine cluster membership, then clustering accuracy is improved, but processing time increases significantly

Engineering Contradiction:
Improveclustering accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent pre-calculates and stores the density values for all data points before the clustering process begins. This preliminary computation of density information eliminates the need for repeated density calculations during iterative cluster membership determination, thereby maintaining clustering accuracy while significantly reducing processing time.

Inventive Principle:
Principle #10Preliminary action

2Ease of operation

If multiple parameters are required for clustering, then clustering control is improved, but algorithm complexity increases

Engineering Contradiction:
Improveclustering controlVSAvoidalgorithm complexity
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The patent extracts and utilizes only the essential density information from the data, ignoring less critical features. By focusing solely on density-based clustering with a single cutoff parameter, the algorithm achieves effective clustering control while maintaining simplicity and avoiding the complexity of multi-parameter approaches.

Inventive Principle:
Principle #2Taking out (Extraction)

3Stability of the object's composition

If cluster shape is constrained, then clustering consistency is improved, but ability to handle varying densities deteriorates

Engineering Contradiction:
Improveclustering consistencyVSAvoidhandling varying densities
Core Design Contradiction:
Stability of the object's compositionVSAdaptability or versatility

Solution Approach 1:

Instead of constraining data to fit predefined cluster shapes, the patent inverts the approach by allowing clusters to naturally form according to the density distribution of the data. This density-driven approach maintains clustering consistency through the cutoff parameter while simultaneously adapting to varying densities and arbitrary cluster shapes.

Inventive Principle:
Principle #13The other way round (Inversion)

Data Source

PatentEP2504776B1Density based clustering for multidimensional data
Publication Date: 2019.06.26 ZYMEWORKS BC INC
  • EP2504776B1 patent drawingFigure 1
  • EP2504776B1 patent drawingFigure 2
  • EP2504776B1 patent drawingFigure 3

AI summary

A new density based clustering method for clustering data points in multidimensional space is described. Each point has a neighborhood consisting of all points that are within a preset cutoff radius or distance. Each point is assigned a density measure based on the number of points in its neighborhood. Any point that has a higher density than any of its neighboring points is the centre of a cluster and is assigned a unique cluster ID. Every other point follows a path through the graph of neighboring points such that density is increasing as fast as possible until a cluster center is reached. The algorithm's performance is demonstrated on a one-dimensional, two-dimensional, and 18-dimensional dataset.