Selecting representative metrics datasets for efficient detection of anomalous data
US10009363B2 · kind B2 · utility
Assignee
Inventors
Key dates
| Filing date | Jun 9, 2016 |
| Grant date | Jun 26, 2018 |
| Priority date | — |
| Expiry date | Dec 31, 2036 |
Classification
- Technology area (CPC H)Electricity
- CPC primaryH04L63/1416
- WIPO fieldDigital communication
- WIPO sectorElectrical engineering
Abstract
Certain embodiments involve selecting metrics that are representative of large metrics datasets and that are usable for efficiently performing anomaly detection. For example, a data graph is generated that represents metrics datasets having values for respective metrics. Each node in the graph represents a respective metric, and each edge between nodes represents a respective correlation between a given pair of the metrics datasets. The nodes are grouped into clusters. For each cluster, a principal component dataset is determined and a representative metric is selected using the principal component dataset. A principal component dataset is a linear combination of metrics datasets (or standardized versions of the datasets) represented by a cluster. The representative metric for each cluster is the metric whose dataset was the greatest contributor to the principal component (e.g., the most heavily weighted metric in the linear combination). An anomaly detection is performed on the selected representative metrics.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.