Time Series Classification via Shapelet Learning and Difficulty Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing time series class classification methods face challenges in accurately labeling data due to noise and human error, especially in distinguishing between normal and abnormal classes, which affects classification performance and interpretation difficulty.
Innovation Solution
An information processing device that classifies time series data into classes using shapelet learning, updates partial waveform patterns by fitting them to time series data, and recalculates classification and interpretation difficulty levels to improve accuracy and correct misclassifications, utilizing a system with an input unit, classification units, and difficulty level update mechanisms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If shapelet learning is used to classify time series data, then classification performance is improved, but interpretation difficulty increases due to noise and human labeling errors
Solution Approach 1:
The system calculates interpretation difficulty levels for each time series data point and uses this information to iteratively reclassify data. High difficulty cases are identified and can be reviewed or re-labeled, creating a feedback loop that improves overall classification accuracy while maintaining interpretability through difficulty-based prioritization
Solution Approach 2:
Instead of requiring perfect classification of all data points, the system focuses on improving classification by updating shapelets and recalculating difficulty levels for problematic cases. This partial action approach targets specific difficult cases rather than attempting to optimize all data simultaneously, reducing interpretation difficulty while maintaining classification performance
2Loss of information
If human labeling is used for class classification, then classification basis can be clarified, but mislabeling occurs due to noise and human error
Solution Approach 1:
The system enables self-correction of labeling errors by automatically calculating interpretation difficulty levels and using these to identify potentially mislabeled data points. The iterative reclassification process allows the system to self-correct human labeling errors without requiring constant external verification, improving reliability while preserving the classification basis provided by human experts
Solution Approach 2:
The system implements feedback mechanisms where classification results and difficulty levels are used to identify potential mislabeling. Cases with high interpretation difficulty are flagged for review, creating a feedback loop that detects and corrects human labeling errors while maintaining the valuable classification basis that human experts provide
3Measurement precision
If iterative updating of shapelets is performed, then classification accuracy is improved, but processing time increases
Solution Approach 1:
The system performs iterative updating of shapelets and reclassification, but focuses computational effort on data points with high interpretation difficulty levels. By partially processing only the most problematic cases in each iteration rather than all data, the system improves classification accuracy while minimizing the time loss associated with iterative processing
Data Source
AI summary
An information processing device includes processing circuitry configured to classify a plurality of partial waveform patterns that characterize a plurality of time series data into a plurality of classes based on the plurality of time series data classified into the plurality of classes, update shapes of the partial waveform patterns by fitting the partial waveform patterns to the time series data of the corresponding class, and reclassify the plurality of time series data into the plurality of classes based on the updated partial waveform patterns and difficulty levels that represent degrees of difficulty of classification and interpretation of the time series data.