Multi-variate Time Series Clustering via Change-point Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems fail to efficiently segment multi-variate time-series datasets due to variance across features, leading to inaccurate and inefficient segment length calculations, as techniques designed for single features cannot be directly applied to multi-variate datasets.
Innovation Solution
The method involves segmenting each feature at transition points, clustering segments into clusters with similar lengths, selecting a representative segment length for each cluster, and applying a threshold test to determine a sufficient segment length for the entire dataset, generating a homogenous feature subset for multi-variate change-point segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If transition points are identified for each feature individually, then segmentation can be performed on multi-variate time series data, but the variance across features leads to inaccurate and inefficient segment length calculations
Solution Approach 1:
The patent segments each feature individually at its transition points, then clusters these segments into groups with similar lengths. This segmentation approach allows the system to handle multi-variate time series data by breaking down the complex problem into manageable parts - processing each feature separately while then integrating results through clustering, thereby achieving accurate segment length calculations without being overwhelmed by feature variance
Solution Approach 2:
The patent introduces a clustering parameter to group segments by similar lengths, and uses this clustered information to determine a unified segment length for the entire dataset. By changing the approach from direct individual feature processing to a two-stage process (segmentation followed by clustering-based parameter determination), the system resolves the accuracy issue caused by feature variance while maintaining computational efficiency
2Ease of operation
If a unified segment length is applied to all features, then processing is simplified, but it cannot account for variance across different features
Solution Approach 1:
The patent segments each feature at its specific transition points, creating feature-specific segments. These segments are then clustered by length, and a representative segment length is selected for each cluster. This segmentation approach allows the system to capture feature-specific characteristics while ultimately producing a unified segment length that works across all features, balancing simplicity and accuracy
Solution Approach 2:
The patent creates a universal segment length determination method that works across all features in the multi-variate time series dataset. By using clustering to find common patterns in segment lengths across different features, the system derives a unified segment length that can be applied universally to the entire dataset, achieving both processing simplicity and segmentation accuracy simultaneously
3Loss of information
If features are segmented at all transition points, then critical information is captured, but excessive sensitivity leads to inefficient processing
Solution Approach 1:
The patent segments features at transition points that meet a predetermined threshold, ensuring critical information is captured. These segments are then clustered by length, and a representative segment length is selected for each cluster. This segmentation approach with threshold-based transition point identification ensures important information is captured while avoiding excessive sensitivity that would lead to inefficient processing
Solution Approach 2:
The patent uses a predetermined threshold parameter to filter transition points, ensuring only significant changes are captured as segments. This threshold parameter acts as a filter that maintains information capture effectiveness while preventing excessive segmentation that would reduce processing efficiency. The clustering step further optimizes this by selecting representative segment lengths that balance information completeness with processing efficiency
Data Source
AI summary
One embodiment provides a method, including: receiving a multi-variate time-series dataset comprising a plurality of time-dependent datasets; for each of the plurality of time-dependent datasets, segmenting each of the plurality of time-dependent datasets at a transition point; clustering segments of the plurality of time-dependent datasets into clusters having similar lengths of segments; for each cluster (i) selecting a representative segment length and (ii) identifying a feature subset in that cluster; identifying, across the feature subsets, subset transition points, wherein each of the subset transition points corresponds to a change in value that meets a predetermined threshold within its corresponding feature subset; and determining, by applying a threshold test to the subset transition points, a segment length to be used in segmenting the entire multi-variate time-series dataset.