Discovering dis-associations between periods of time-series


I'm interested in discovering some kind of dis-associations between the periods of a time series based on its data, e.g., find some (unknown number of) periods where the data is not similar with the data from another period.

Also I would like to compare the same data but over 2 years (something like DTW?).

I get my data Excel as a two-column list:

c1=date (one per each day of the year), c2=Data To Analyze

So, what algorithms could I use and in what software?

Update/Later edit: I'm looking for dates as cut-off points from which the DataToAnalyze could be part of another cluster of consecutive dates. For example:

2014-1-1 --> 2014-3-10

are part of Cluster_1 based on DataToAnalyze. And:

2014-3-11 --> 2014-5-2

are part of Cluster_2 based on DataToAnalyze, and so on. So, clusters of consecutive dates should be automatically determined based on some algorithms, which is what I'm looking for. Which ones (or which software) would be applicable to this problem?


Posted 2014-09-30T19:53:20.163

Reputation: 121

Hi Catalin, and welcome to DS! As it stands, your question is hard to answer. What does "not similar" mean? Are you looking to detect structural changes, like shifting levels? Outliers? Changes in seasonality? What is the "data" (any number, or counts, lots of zeros or few)? Please consider editing your question. – Stephan Kolassa – 2014-10-01T09:50:04.953

i've updated the question – Catalin – 2014-10-01T15:30:48.633

No answers