Data for Evaluation of Stream Data Analysis Algorithms

This collection of datasets have been generated with MDCStream for evaluating clustering and outlier detection algorithms in stream data analysis. The temporal behavior is assumed in the order in which data points appear in the file, this means: simultaneity is not considered and the time-difference between consecutive data points (i.e., consecutive rows) is the unit. Datasets are arranged in 9 folders according to the data challenge: base (baseline), nonstationary (clusters coexist, appear and disappear randomly), sequential (clusters happen sequentially), moving (cluster centroids move as time passes), medium-outliers (outliers account for 5% to 15% of the data), many-outliers (outliers account for 15% to 40% of the data), close (the space is reduced and clusters are very close each other), density-differences (distributions underlying point generation are highly varied), overlap (the data generation favors cluster overlap).