Seasonal Pattern Classification via Signal Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current time series forecasting models, such as Holt-Winters, primarily focus on generating forecasts and provide limited information on seasonal patterns, with seasonal indices not being easily interpretable outside of the forecasting context and failing to account for both dense and sparse patterns effectively.
Innovation Solution
The approach involves detecting and classifying seasonal patterns by analyzing time series data to identify recurrent highs and lows, decomposing data into dense and noise signals for independent processing, and using supervised or unsupervised methods to classify patterns, generating summaries that highlight stretches of time associated with seasonal classes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Holt-Winters forecasting algorithm is used to generate forecasts, then forecasting accuracy is improved, but interpretability of seasonal patterns deteriorates
Solution Approach 1:
The patent extracts seasonal patterns from the Holt-Winters forecasting algorithm by separating the seasonal index calculations from the forecasting process. The seasonal patterns are identified and characterized independently, allowing them to be interpreted and visualized separately from the forecast generation, thus resolving the contradiction between maintaining forecasting accuracy and improving pattern interpretability
Solution Approach 2:
The patent introduces an intermediary layer that processes the seasonal indices generated by Holt-Winters. This intermediary layer transforms the internal seasonal index data into meaningful, interpretable patterns through clustering and classification, bridging the gap between the forecasting algorithm's internal representations and human-understandable seasonal patterns
2Productivity
If seasonal indices are used internally in forecasting models, then forecasting capability is improved, but ease of interpretation outside forecasting context deteriorates
Solution Approach 1:
The patent segments the seasonal analysis into distinct components: identification of seasonal periods, extraction of seasonal patterns, classification of pattern types, and generation of interpretable summaries. This segmentation allows the seasonal indices to be processed and presented in a way that is independently interpretable without requiring the full forecasting context
Solution Approach 2:
The patent transforms the internal seasonal index parameters into different representations that are more suitable for interpretation. By changing the parameters from raw seasonal indices to classified pattern types and stretch summaries, the system maintains forecasting capability while improving ease of interpretation for end users
3Device complexity
If traditional seasonal models are applied, then computational simplicity is improved, but ability to account for both dense and sparse patterns deteriorates
Solution Approach 1:
The patent introduces dynamic adaptability by detecting whether seasonal patterns are dense or sparse and adjusting the analysis approach accordingly. The system dynamically identifies the appropriate seasonal period and applies suitable classification methods, allowing it to handle both dense and sparse patterns effectively without requiring complex predefined configurations
Solution Approach 2:
The patent implements self-service through automated detection and classification of seasonal patterns. The system automatically identifies seasonal periods, extracts patterns, and classifies them without requiring manual configuration or intervention, thereby maintaining computational simplicity while improving the ability to handle diverse pattern types
Data Source
AI summary
Techniques are described for classifying seasonal patterns in a time series. In an embodiment, a set of time series data is decomposed to generate a noise signal and a dense signal, where the noise signal includes a plurality of sparse features from the set of time series data and the dense signal includes a plurality of dense features from the set of time series data. A set of one or more sparse features from the noise signal is selected for retention. After selecting the sparse features, a modified set of time series data is generated by combining the set of one or more sparse features with a set of one or more dense features from the plurality of dense features. At least one seasonal pattern is identified from the modified set of time series data. A summary for the seasonal pattern may then be generated and stored.