Singular Spectrum Analysis (SSA) is a relatively new approach to modelling time series data. It is a novel and powerful technique applicable to many practical problems such as the study of classical time series, multivariate statistics, multivariate geometry, dynamical systems and signal processing (Hassani, 2007). The SSA method of time series analysis applies nonparametric techniques to decompose time series into principal components. It is particularly valuable for long time series, for which patterns are difficult to visualize and analyze.
Let me know from the research community, what freely available software/or code they use for SSA?
Eou may use CaterpillarSSA 3.40 software on http://www.gistatgroup.com/gus/programs.html. This page is in Russian, but there is the textbook in English on http://www.amazon.com/exec/obidos/ASIN/1584881941.
The point how the Caterpillar or SSA in general performs better than "classic" Karhunen-Loeve Transform still escapes me.
On the Biology side (you have placed Bioinformatics topic there), close to the soil, the practical questions are more often formulated not as prognosis based on a given time series (complete or with omitted observations). Typically you want to know, based on a time series of estimated gene expression (microarray, RT-PCR, RNAseq, etc.) if the gene is rhythmically expressed, what is the period, amplitude and phase of this baseline periodicity. You also want to attach some estimation of confidence in labeling genes periodic. Also, what they call "short noisy time series" in economics or climate data analysis would be an exorbitant opulence in Functional Genomics. Here you can count time points with your fingers and the signal sinks in noise like pebble splash in hurricane. There are many methods proposed to deal with this kind of time series, I can recommend some of mine:
http://www.ncbi.nlm.nih.gov/pubmed/17118131
http://www.ncbi.nlm.nih.gov/pubmed/17571920
Two papers with description of functionality and code examples of Rssa (http://cran.r-project.org/web/packages/Rssa/index.html) mentioned by @Xuanlong Ma can be helpful:
http://arxiv.org/abs/1206.6910
http://arxiv.org/abs/1309.5050
Is it possible to use singular spectral analysis for natural variability of a long time observed climate data? Moreover, is it available in Matlab?
Muhammad Ejaz Tanveer, you can look at this paper http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0107222 ("Application of the Singular Spectrum Analysis Technique to Study the Recent Hiatus on the Global Surface Temperature Record")
However, they use Rssa.
To Ms. Golyandina,
Ive been using your RSSA package for over 6 months but i cant seem to use it if the input is a daily gridded dataset... Just hoping if you can help me on this.
Many thanks!!!
Warren Luzano, do you mean 3D (2D spacial + 1D temporal) data in a special format? Did you try to convert the data to an appropriate format, which can be processed by Rssa (Rssa is able to process 3D data (array))?
The special format you're referring is NETCDF. Been using RSSA for several months but cant seem to re-create a similar paper by Karmakar (whom I had asked permission and he said its ok).
He applied MSSA, but used matlab instead.
Mam, If you have the time to browse and see if its possible to use RSSA to come up with the outputs . Thank you so much ...
Link: http://iopscience.iop.org/1748-9326/10/5/054018/media/erl054018_suppdata.pdf
Kspectra is the software with good tutorials, available for Macintosh
Dear Khalid,
You may find a matlab code at
https://github.com/dfourer/ASTRES_toolbox/blob/master/SSA/ssa.m
best regards,
Dear Khalid,
I use AutoSignal software, but is not for free.
You can find an extended SSA application on my recent paper:
DOI: 10.1186/s40645-018-0236-z
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence