One point to take into account is that these datasets do benchmark against known attacks and do not measure the capability of detection against new attacks.The other thing is that if a dataset includes benign traffic it will correspond to a specific user profile behaviour. This implies that one has to be very careful on the type of conclusions that one draws on these datasets.
Even though, there were several bench mark data sets available to test an anomaly detector, the better choice would be about the appropriateness of the data and also whether the data is recent enough to imitate the characteristics of today network traffic.
Anomaly Detection: Algorithms, Explanations, Applications, Anomaly Detection: Algorithms, Explanations, Applications have created a large number of training data sets using data in UIUC repo ( data set Anomaly Detection Meta-Analysis Benchmarks & paper, http://www.outlier-analytics.org...).
KDD cup 1999 dataset ( labeled) is a famous choice. However, unlike many real data sets, it is balanced. KDD Cup 1999 Data
Data sets at http://odds.cs.stonybrook.edu/ are pretty good.
Kaggle has a credit card fraud data set: https://www.kaggle.com/dalpozz/c... (Credit Card Fraud DetectionAnonymized credit card transactions labeled as fraudulent or genuine)
Breast Cancer dataset https://archive.ics.uci.edu/ml/d... ( labeled)
https://github.com/numenta/NAB - OK, but generated data set