What is it about?
Outlier detection is one of the most popular tasks among data scientists because outliers might have to be removed from datasets due to making it low accuracy in prediction tasks, for example. However, in case of multivariate time series datasets, it is difficult to find them. Therefore, many algorithms have been developed so far. This study shows a new algorithm called Unsupervised Feature Extraction using Kernel and Stacking (UFEKS), which constructs a kernel matrix using RBF kernel from each time series and horizontally concatenating all the matrices. Its row vectors represent features of multivariate time series. This study also shows an effectiveness of the algorithm with the experimental results using real-world datasets.
Photo by Chris Liverani on Unsplash
Why is it important?
In recent years, Internet of things (IoT) devices like sensors of temperature, humidity, and pressure, are installed in many systems and sometimes those collected data are stored in cloud systems. Therefore, we can easily use the data remotely for a variety of services, however, it is not so easy tasks to handle them because those data consist of multivariate time series and we must take both time-wise and variable-wise association into account. To address the issue, the new algorithm called UFEKS has been developed.
Read the Original
This page is a summary of: Unsupervised feature extraction from multivariate time series for outlier detection, Intelligent Data Analysis, November 2022, IOS Press, DOI: 10.3233/ida-216128.
You can read the full text:
The following have contributed to this page