What is your goal ? it largely depends on it . What do you want to "learn" from network topologies ? 

Hi Mohamed,

The selection of a "Deep" or conventional machine learning technique/algorithm very much depends on either trial and error, or the specific "type" of data and problem you are trying to solve. For example, "Deep" learning may lend itself to custom, optimized neural network implementations to classify pictures of cats. However, if your problem is associated to training a general classifier where you want to create feature selection for training, then there are many methods such as decision tree, gradient boosting, random forest, logistic regression, k nearest neighbor, case-based reasoning, k-means clustering, naive bayes.... etc .... etc. You first have to articulate properly what is your problem, then compare the methods of statistical analysis other research has undertaken in solving a similar problem to yours. 

For clarity, you do not necessarily require a "deep" learning approach if you simply was to creating a trainer for classifying data based on labels produced from a dataset. What machine learning algorithm you try, then tune and adjust for your task is an arbitrary choice where different types perform better or worse for different data and problems.

Here is an excellent free online book that discusses the concepts of machine learning:

http://alex.smola.org/drafts/thebook.pdf

I hope this helps in some way.

Thank you Mr Sashank,

The problem that I aim to solve is the recognition of the behavior patterns of Internet users based on their connection data.

Hi Ryan thank you for your answer and for the book,

As I sad to Mr Sashank the problem that I aim to solve is the recognition of the behavior patterns of a Cloud service users based on their connection information which are captured and stored in data set.

These data sets are very large and complex (some of them are structural and other doesn't). This because they are captured over thousand of Cloud service and they contain many information of Cloud users such as users IP addresses, Port number, protocol, etc. This are the reasons for which I have thinked in deep learning techniques.

But I still don't know which technique is pretty convenable to this problem of behavior patterns recognition..

The term "deep learning" is still vague. And besides with the type of data you said you have, I get the nod you are simply trying to do statistical machine learning. Because you have huge and complex data doesn't qualify your problem for deep learning.

Given the data that you have, I would like to know if it labelled (supervised), or unlabelled(unsupervised)? i.e. if each observation in your dataset contain a field that classifies it? Anyways, the Internet is awash with books, tutorials and codes that can aid you in doing machine learning.

If you want to understand basic theory (not much math) and practicality (codes in R language) of various learning methods, you should read this book "An Introduction to Statistical Learning with Applications in R", link provided below. It contains straight descriptions of the methods, including codes and datasets to handle different types of Pattern Recognition. 

Another book is that Pattern Recognition and Machine Learning, included below, a bit heavy on the theory.

Beside R, there are machine learning libraries for Python (scikit-learn) and other popular languages.

http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf

http://scikit-learn.org/stable/

http://www-bcf.usc.edu/~gareth/ISL/

Hi Mr Olumuyiwa thank you for sharing the book and so much for the tutorials,

The data that I have are labelled, but the learning seem to be unsupervised because I aim to perform some behavior patterns recognition based on these data. For example, in the first step I would like to extract from the collected data sets the most behavior patterns that some Internet users follow.

For example, we have a web site that sell clothes and there are clients in all over the world, my goal is to know the most clothes that peoples of each country buy. In other words, I would like to get the most common behavior patterns of each nationality.

However, I don't have any ideas about the behavior patterns of clients. This is the reason which make me thinking about unsupervised learning.

But, what is the best technology to perform this kind of behavior recognition?

Oh well, then you may try Clustering, especially Hierarchical Clustering. Those books have dedicated chapters for doing this.

By the way, please upvote my answers since you found them useful.

Thank you.

OK Mr Olumuyiwa I will try Clustering.

Yes your answers are very useful.

Thank you.