An Anomaly Detection Algorithm Selection Service for IoT Stream Data Based on Tsfresh Tool and Genetic Algorithm

Table 3

The datasets used in our experiment.

Name	Source	Number of time-series	Number of time stamps	Ratio of anomalous data (%)	Characteristic
Dataset 1	Yahoo [18]	100	1680	0.5	Artificial univariate time-series data comprises of anomalies’ change point where it changes the mean of the time series
Dataset 2	Yahoo [18]	100	1680	0.3	Artificial univariate time-series data with anomalies and seasonality are introduced at random points
Dataset 3	Yahoo [18]	100	1421	0.3	Artificial univariate time-series data
Dataset 4	Yahoo [18]	67	1420	1.9	A univariate Yahoo services time-series dataset recording the traffic in which anomalies are by-hand pigeonholed. Majority of the time-series are static
NYCT	NAB [17]	1	10320	0.05	A univariate New York City taxi request time-series dataset comprising the New York City (NYC) taxi demand from July 1, 2014, to January 31, 2015, with an observation of the no. of passengers noted down every half hour. It comprises five shared anomalies that arise in the NYC: Christmas, thanksgiving, marathon, snowstorm, and New Year’s Day.