I have a dataset that only has two columns, ID and DATA, and 100 rows. Below is the example of how my dataset looks like:

ID     DATA  
---    ----------------  
1      ac 09 bb 46  
2      4f cd e2  
3      ae bc  
1      ac 09 bc 46  
2      4f ce e2  
3      ae bd  

I would like to setup up algorithms for detecting an anomaly in time series, and I plan to use time series while at the same time data classification for that:

I've wondered whether:

  1. I can develop Time Series model on 'ID' column to measure frequencies between the IDs in the flow, and at the same time I build a classification model on 'ID' and 'DATA' column, where 'DATA' will be classified based on categorical label 'ID'?

  2. Are there any techniques that could parallelize training set data? I could train a time series model first, and then train a classifier to classify them based on their classes/labels, but I'm just thinking that if there's an algorithm that could train both in parallel.

  3. I assumed that my case is multivariate time series. I was wondering how to handle spaces in the 'DATA' column?

  4. Can I just sum up the errors in prediction from both models to indicate anomaly? How can I use these errors to build an anomaly detection model?

  • $\begingroup$ Hi Ana, welcome to CV! Note that a part of this site's philosophy is to focus only on the question, so phrases like "I am new to machine learning", "Any help is appreciated" or "Thanks" don't belong here. Although it might feel rude at first, it is implied that you are happy and thankful for getting help. Consider taking the tour or checking How to ask to get introduced. $\endgroup$ – Jan Kukacka Apr 24 '18 at 10:13
  • $\begingroup$ what is time series here? is it temporal data? What does ID and data column represents? $\endgroup$ – Arpit Sisodia Oct 4 '18 at 14:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.