CatSight, a direct path to proper multi-variate time series change detection: perceiving a concept drift through common spatial pattern

Abstract Detecting changes in data streams, with the flowing continuously, is an important problem which Industry 4.0 has to deal with. In industrial monitoring, distribution may vary after a change machine’s operating point; this situation known as concept drift, and it key detecting change. One drawback of conventional machine learning algorithms that they are usually static, trained offline, require monitoring at input level. A data, relationship between output would result deterioration predictive performance models due lack ability generalize model new concepts. Drift methods emerge solution identify drift data. This paper proposes approach for detection—a novel sudden or abrupt most common found processes-, called CatSight. Briefly, method composed two steps: (i) Use Common Spatial Patterns (a statistical streaming, closely related Principal Component Analysis) maximize difference different distributions multivariate temporal (ii) Machine Learning detect whether flow been occurred not. The CatSight method, evaluated on real use case, training six state art (ML) classifiers; obtained results indicate how adequate is.