Specifically, I want to know real-time machine learning models that are capable of identifying anomalies considering streams of data with multiple features (i.e. multivariate time series).
I have found a scoring mechanism called "Numenta" which scores (NAB) real time machine learning models. However most of the models compared there are for univariate time series and have not been extended for multivariate cases. Therefore, I am searching for similar real-time machine learning models that can handle multivariate input data streams.
See the link. It is the standard work in the field. Best, D. Booth
http://www-bcf.usc.edu/~gareth/ISL/
Jerome Dinal Herath There are various models that you can use for real-time anomaly detection. Since, its real-time my suggestion would be to use supervised learning. Some of the models that you can use are one class svm, random forest, auto encoders, KNN, and decision tress. If you are using regression, then you need to have a limit to determine if its an anomaly else you can do 2 class classification. You can train the model and use it for real time scoring. I am not sure about the latency of your real time data but in any case above should work. I personally have used auto encoders for real-time scoring with a latency of couple of milli seconds.
If you don't have labels for your data then see this paper for unsupervised model
Amruthnath, Nagdev, and Tarun Gupta. "A research study on unsupervised machine learning algorithms for early fault detection in predictive maintenance." In 2018 5th International Conference on Industrial Engineering and Applications (ICIEA), pp. 355-361. IEEE, 2018.
- Badges
- Science topic