I am working on anomaly detection in real-time and my data looks attached and it's unlabelled. I want to detect if the next minute timestamp is an anomaly or not. My main column in num_part which is number of players and all other columns are supporting metrics.
Issue:-
1.) I have some kind of data for 50 different domains.
2.) There is differnet kind of seasonality in different domains like on WednesdayTuesday domain 1 goes for Maintainance on tuesday domain b etc.
The approach I am using:-
As of now, I am calculating Z-score for every five-minute interval and comparing it with historical data and alerting but getting lots of false positive.
I am even considering it a streaming data like we have in sensors and not a time series.
I want to alert my team in real-time if the incoming next data point is an anomaly.
https://stats.stackexchange.com/questions/452988/anomaly-detection-in-unlabelled-multivariate-minute-by-minute-time-series-data
From a data science point of view, you may apply a compression rate on your data provided that you organize your data properly. Then, you just check the compression rate variation instead of the data variation itself. I have already used this method successfully for anomaly detection on sensor data. Again, please make sure you make proper data first and apply a relevant pre-filter to reduce false alarms before having your compression algorithm.
You may find the paper here:
Article Edge analytics for anomaly detection in water networks by an...
Reza Fotohi that’s my question only stack exchange and it doesn’t have any answer. Not sure why you are sharing it with me.
- Badges
- Science topic
- Similar topics
- Analytical Chemistry
- Column