I have came across a binary classification problem which I dont think falls under traditional binary classification. So just wanted to share with you guys to get some more ideas.

I have a series of data records which is in a chronological order. For simplicity lets say I have 100 data records and I have a target value as 1 or 0 for every consecutive 10 data values.

e.g. data = [1,2,3,4,5,6,7,8,9,10]

Training data would look like this:

[1,2,3] => 1, [2,3,4] => 0, [3,4,5] => 0, [4,5,6] => 0, [5,6,7] => 1 and so on and so forth. please note that order of values does matters

I am currently struggling on how to handle this idea. Is there any Machine learning tools that can handle this? like Amazon Machine Learning or BigML etc.

My approach:

I am thinking of using some kind of hash function that can hash the series to specific value and that value can be used for one to one binary classification. However I am not sure if I am going in the right direction.

EDIT: More precisely I have a data consisting of 5 numerical features. Each instance of these 5 feature values are coming in a timely manner like every second. This instance or vector of 5 feature values independent of each other does not have any a significance however when consider in a group they have (In this case group is consecutive 100 vectors). These groups can be labeled as either 1 or 0 based on their characteristics. Now I wanted to build a classifier for this that can in future classify the data series of 100 instance as 1 or 0.

Any help would be appreciated.

  • $\begingroup$ I can't see what's so unusual about this problem, really. Your features would be the first, second, and so on values in the series. $\endgroup$ – Firebug Sep 28 '16 at 19:42
  • $\begingroup$ In general, this problem is called time series classification. $\endgroup$ – Sycorax Sep 28 '16 at 20:40
  • $\begingroup$ @Sycorax: "time series classification" could also refer to classifying time series, which is a different problem than time-ordered classification data. $\endgroup$ – Stephan Kolassa Sep 29 '16 at 9:39
  • $\begingroup$ What is your goal? Do you want to forecast future classifications? Do you have additional explanatory data (time series or other)? Plus: if you already know that each block of ten observations will have the same classification, then you could simply take every tenth data point without losing any information. Is there any specific reason why you don't want to do that? $\endgroup$ – Stephan Kolassa Sep 29 '16 at 9:40
  • $\begingroup$ @StephanKolassa I'm not clear on the distinction you're making. Could you please clarify? $\endgroup$ – Sycorax Sep 29 '16 at 13:37

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.