Java API for Automatic Regression (AR), ARIMA, Time Series Analysis

Question

I am looking for either an Opensource or a Free Java API for time series analysis using AR, ARIMA, etc. I need this api to analyze DDOS Attack. I googled around and found 2 solutions, but both do not completely solve the problem:

1) The same question was asked earlier in https://stackoverflow.com/a/4642222/ ... and a solution was published regarding SuanSu Api , but this API is not free

2) Apache Math Library , but this API provides other forms of regression, such as Simple, OLS, GLS, etc., but not automatic regression.

I checked the options in the Machine Learning apis as Mahout, but have failed so far. Please suggest a suitable API

Answer 1

I spent the 4th year on a Computing project to implement time series forecasting to predict the use of Java heaps using ARIMA, Holt Winters, etc., so I could help you with that.

Your best option is to use the R language, you can call the prediction libraries provided by R through Java using the JRI library found here . R is well documented, free and open source. You can even run R on the server and then make calls using the command line using Rserve, which then returns HTTP forecasts, but the JRI is the local equivalent if the memory serves me properly.

If you have any questions, let me know.

Answer 2

Check out the spark-timeseries . The source code is mostly Scala, but it is relatively easy to use libraries from Java. If you are in the place where you are doing time series analysis on the JVM, you should consider learning Spark / Scala anyway.

The library is young at the time of this writing and has room for improvement and growth, but since version 0.3 it implements AR, ARIMA, simple exponential smoothing (EWMA) and Holt-Zithers smoothing. Areas for improvement are ARIMA's best automatic algorithm, seasonal ARIMA support, and space-based state modeling, but this is already very useful.