Time series analysis using a clustering based symbolic representation
US10248713B2 · kind B2 · utility
Assignee
Inventors
Key dates
| Filing date | Nov 30, 2016 |
| Grant date | Apr 2, 2019 |
| Priority date | — |
| Expiry date | Aug 17, 2037 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06N20/20
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
Techniques are described for performing a time series analysis using a clustering based symbolic representation. Implementations employ a clustering based symbolic representation applied to time series data. In some implementations, the time series data is discretized into subsequences with regular time intervals, and symbols encoding the time intervals may be derived by performing clustering algorithms on the subsequences. In the new representation, a time series is transformed into a sequence of categorical values. The symbolic representation is suitable to perform time series classification and forecast with higher accuracy and greater efficiency compared to previously used techniques. Through use of the symbolic representation, a dimension reduction is applied to transform the time sequences to a feature space with lower dimensions. As output of such transformation, a new representation is obtained based on the original time series. This new reduced-dimension representation improves the efficiency of time series data mining and forecasting.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.