Dear Azedine, I suggest you the following link (from Pisa University, Italy), where you can find methods and ideas about big data analysis on today-matters.  Gianluca

http://www.sobigdata.eu/

Dear Azedine,

first you should tell what you understand under Big Data and Big Data Analytics.

If you consider the volume aspect of Big Data and mean a data set with more cases than features, than I doubt that Big Data Analytics is necessary at all. Usually, if the data are large they are cut down or aggregated for analysis purposesinto handy chunks. The usual statistical and ML techniques where developed for small 'til medium sized data sets and should generalize generalize to larger data sets. A pretty argument about Big Data follows the quote of Gelman in http://nbviewer.jupyter.org/github/CamDavidsonPilon/Probabilistic-Programming-and-Bayesian-Methods-for-Hackers/blob/master/Chapter1_Introduction/Ch1_Introduction_PyMC2.ipynb

If you define Big Data as containing more features than cases, than you are talking about approaches to high-dimensional data for which specialized techniques exist. Search e.g. for "high-dimensional statistics" or "high-dimensional probability". But in this case you are usually goint to figure out which features are relevant and which are not, in order to avoid spurious correlations. For that their are PCA, ICA and LDA.

If you consider the variety aspect of Big Data (structured, semi-structured vs. unstructured information), than clearly unstructured information like images, text or time series, require specialized analytical techniques. So in that case you need to specify what are the data you like to process.

If you consider the velocity aspect than clearly real-time data require different analytical approaches than static data.

However, if your problem is how to process large amounts of data, the question is independent of any analytical approach. In that case I find approximative approaches like MinHashing, Locality Senstive Hashing etc. extremly useful.

And why does it need to be new? The problem of coping with big data (i.e. data too large to fit into main memory) are not new and already in the 70th and 80th approaches where developed (like hashing etc.) to cope with this question which are still useful.

Thanks Thomas ...

Complete and perfect response