MDCStream

Abstract

The establishment of modern technological paradigms like ubiquitous computing, big data, cyber-physical systems, or communication networks has strongly increased the need for efficient, effective data stream analysis. MDCStream is a MATLAB tool for generating temporal-dependent numerical datasets in order to stress-test stream data classification, clustering, and outlier detection algorithms. MDCStream is built on MDCGen, therefore showing a high flexibility for creating a wide diversity of data scenarios. To show an example of the potential of MDCStream, we tested a stream data clustering algorithm recently proposed in the literature with datasets generated with MDCStream. Datasets were designed to draw challenges related to space geometries and concept drift.