Shapelet Learning for Time Series AUC-Optimized Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to optimize classification models for time series data sequences by optimizing performance indicators like pAUC or AUC, and do not concurrently learn feature waveforms as evidence for prediction.
Innovation Solution
A method that generates feature vectors based on distances between time series data sequences and feature waveforms, updates the feature waveforms using a performance indicator parameter and model parameter, including weights, to optimize the classification model for time series data analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional classification methods are used for time series data, then the classification model can be built, but the performance indicator (pAUC or AUC) cannot be optimized
Solution Approach 1:
The patent transforms the classification model learning problem into an optimization problem by changing the objective parameter from conventional accuracy metrics to pAUC or AUC. The learning apparatus adjusts model parameters to directly optimize these performance indicators through gradient-based optimization methods, enabling precise control over false positive rates and improved anomaly detection performance.
Solution Approach 2:
The patent replaces conventional classification algorithms with a gradient-based optimization system. Instead of using traditional mechanical classification rules, the system uses continuous optimization with differentiable loss functions (such as ranking loss) to learn model parameters that maximize pAUC or AUC, substituting discrete classification mechanics with continuous optimization processes.
2Loss of information
If feature waveforms are learned concurrently with classification model, then evidence for prediction is provided, but the performance indicator optimization is not achieved
Solution Approach 1:
The patent merges the feature waveform learning process with the classification model optimization into a unified framework. Both the classification model parameters and feature waveform parameters are learned simultaneously by optimizing a single objective function based on pAUC or AUC, allowing the system to provide prediction evidence through feature waveforms while achieving performance indicator optimization.
Solution Approach 2:
The patent creates a universal learning framework that simultaneously performs multiple functions: learning classification model parameters, learning feature waveforms as prediction evidence, and optimizing performance indicators. This multi-functional approach allows a single system to handle both evidence generation and performance optimization without separate processing stages.
3Measurement precision
If false positive rate is narrowed to a small range for pAUC optimization, then prediction accuracy is improved, but the classification model becomes more difficult to train
Solution Approach 1:
The patent introduces a ranking loss function as an intermediary between the classification model and the pAUC optimization objective. This intermediary loss function provides a differentiable and computationally efficient approximation that guides the model training process toward pAUC optimization, making it easier to train while maintaining high prediction accuracy in the specified false positive rate range.
Data Source
AI summary
A time series data analysis method, includes: generating a plurality of first feature vectors including feature amounts of a plurality of feature waveforms, based on distances from a plurality of first time series data sequences to the plurality of feature waveforms, the first time series data sequences belonging to a first class; generating a plurality of second feature vectors including feature amounts of the plurality of feature waveforms, based on distances from a plurality of second time series data sequences to the plurality of feature waveforms, the plurality of second time series data sequences belonging to a second class; and updating the plurality of feature waveforms, based on the plurality of first feature vectors, the plurality of second feature vectors, a performance indicator parameter related to a performance indicator for a classification model and a model parameter including weights on the plurality of feature waveforms.