views:

1990

answers:

4

I need to do some analysis of an arbitrary amount of time series in Java.

Among others i need to be able to use Linear regression, various smoothing techniques, filtering, etc.

I'm not very keen of writing all this from scratch so, do you know of any good Java libraries for these kind of analysis?

Edit:

R- Seems like an good choice. It has a powerful set of methods and has good external packages for doing statistical analysis, you can find functions for Linear regression,autocorrelation, holt-winters filtering etc.

I found an port between R and Java which allows R to run in Java.

This should be sufficient for my needs.

+4  A: 

Take a look at JMulTi and NBB Libraries for Time Series Analysis this is what I found. If it is just about analysis then I would recommend R, which has good packages for time series analysis.

seb
I concur on using R for this sort of thing if possible. It is not trivial to directly link to Java, but you can always write from R and read from Java.
Daniel Nesbitt
I meant, when it is just about analysis and to get analysis results, not when it is needed to use the results in a Java program afterwards. How to make R an Java work together, I don't know.
seb
Well it's kinda essential to be able to implement the result in Java, but i suppose if one could find a way to fetch the R-results to Java, it would be a good idea.
Silfverstrom
oh, and thanks for the answer:)
Silfverstrom
+1  A: 

Weka is a well-known machine learning library in Java, but it does not support time series too well. However, some people have added a little time series support in their modified versions, and you can always ask on the mailing list about how you could achieve some specific time series tasks within Weka. See for example one question from April.

Jouni K. Seppänen
A: 

There is a new Java interface to Seats-Tramo the Bank of Spain model-based package.

Martin
A: 

There's also a time series class in jquantlib.

yziquel