Anomaly Detection Using Multiscale C-LSTM for Univariate Time-Series

Table 1

Samples numbers and abnormal proportion of datasets under different window sizes: Num. is the total number of samples in the dataset, and Prop. is the proportion of abnormal samples in the dataset.

Dataset	Window
	20		40		60		80		100
	Num.	Prop. (%)	Num.	Prop. (%)	Num.	Prop. (%)	Num.	Prop. (%)	Num.	Prop. (%)
A1	93593	4.38	92253	6.89	90913	9.41	89573	11.87	88233	14.36
Ad	9496	2.95	9376	5.97	9256	9.08	9136	12.26	9016	15.53
AWS	67417	0.92	67077	1.85	66737	2.79	66397	3.74	66057	4.69
Known	69428	0.56	69288	1.11	69148	1.66	69008	2.22	68868	2.77
Traffic	15531	1.80	15391	3.61	15251	5.35	15111	7.12	14971	8.91
Tweets	158441	0.44	158241	0.88	158041	1.33	157841	1.77	157641	2.22