I am looking forward to the correct ML/algorithm approach for the below issue.

My target here is to predict the target day of the incoming time series below for a new time series. Also below you can find the form of the train dataset. My goal is to train an algorithm on the various patterns of train datasets' time series (lets assume that in the train dataset we can find all the possible forms of time series [with trend, stationarity etc.]) in order to predict the target value based on the expected behavior of similar time series in train dataset.

Train dataset:

TS name Day1 Day2 ... Day50 Target-Day51
TS 1 5 13 ... 16 12
TS 2 8 18 ... 9 16
... 12 2 ... 13 4
TS 4000 3 7 ... 4 10

Incoming time series:

TS name Day1 Day2 ... Day50 Target-Day51
TS 4001 3 22 ... 48 XX

Any ideas please?

Thank you in advance


1 Answer 1


It depends on the structure of the target values. Are they nominal or ordinal? If the target values are nominal, multi-class classification would be used to predict a single group out of many possible different groups.

One option is to make a simplifying assumption by ignoring the order of observations. That is similar to the bag-of-words model in natural language processing (NLP). Then any multi-class classification algorithm can be used, such as Naive Bayes or Random Forest.

Also, time series modeling can be tried. For example, Hidden Markov Model (HMM) could be used to predict the target day given the sequence of previous values.

If those models do not yield useful predictions, more advanced modeling, such as Long Short Term Memory (LSTM), can be tried.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.