Is it worth it? Comparing six deep and classical methods for unsupervised anomaly detection in time series

Open paper
Benchmark Model Rank Results
anomaly-detection-on-ucr-anomaly-archiveMERLIN#5Average F1: 0.27 ±0.0AUC ROC: 0.51 ± 0.0
anomaly-detection-on-ucr-anomaly-archiveMaximally Divergent Intervals (MDI)#6Average F1: 0.25 ±0.0AUC ROC: 0.66 ± 0.0
anomaly-detection-on-ucr-anomaly-archiveGraph Augmented Normalizing Flows (GANF)#7Average F1: 0.23 ±0.021AUC ROC: 0.63 ±0.009
anomaly-detection-on-ucr-anomaly-archiveTransformer Network for Anomaly Detection (TranAD)#8Average F1: 0.18 ±0.003AUC ROC: 0.56 ±0.003
anomaly-detection-on-ucr-anomaly-archiveAutoencoder (AE)#10Average F1: 0.16 ± 0.013AUC ROC: 0.58 ±0.01
anomaly-detection-on-ucr-anomaly-archiveRobust Random Cut Forest (RRCF)#12Average F1: 0.07 ±0.011AUC ROC: 0.56 ± 0.0019

Rank counts only results with a code link.