@inproceedings {pub3851,
	title = {Large Anomaly Detection in Univariate Time Series:
An Empirical Comparison of Machine Learning Algorithms},
	author = {Sina D{\"a}ubener AND Sebastian Schmitt AND Hao Wang AND Thomas B{\"a}ck AND peter krause},
	year = {2019},
	month = {July},
	abstract = {This paper presents an empirical comparison of state-of-the-art machine learning and statistical methods for anomaly detection in univariate time series. 
In particular, we compare random forests, support vector machines, $k$-nearest neighbour regression, artificial neural networks, long short-term memory networks, Twitter{\textquoteright}s anomaly detection method AdVec and an ARIMA model on publicly available data sets with labeled anomalies.
Since the anomalies are labeled, well-established measures of classification accuracy are applicable for evaluating the performance of the anomaly detection algorithms. 
The results clearly demonstrate that using a window of consecutive input values for predicting the next instance yields results which are comparable to or better than the specific anomaly detection algorithm AdVec. 
Using consecutive gradient values for predicting the next gradient value yields similar results as well. 
In both cases, the generalized extreme studentized test is used for detecting the anomalies. 
This study is the first comparison of these six algorithms on both publicly available data sets for anomaly detection in univariate time series, covering a total of 417 time series.},
	publisher = {Unknown},
	booktitle = {19th Industrial Conference on Data Mining ICDM 2019}
}