Unfortunately the best answer might be to get all the data into the same form. I suspect that will tend to, but not always, reduce the information content a little. Perhaps you could use bagging (various methods of chopping the data into short ranges that are then used as categories), and similar procedures to get all your data into the same form. Then you can treat all of it as factors using various data mining algorithms that work with all the data as categories.  However, you might alternatively take a careful look at your data and find a way to make all of it numerical.

You can use the Exploratory Factorial Analysis or Principal Componeents Analysis.

You can make use of the clustering techniques in data mining, or the use of the Principal Component Analysis Algorithm for multivariant analysis

Dears,

Thanks a lot for your answer. I hope, it will help me to be clear about the ideas.

Wish you have a good day.

Best Regards,

I agree with Lewis Maccarter. The best way is to bring all attributes to the same form. This is because we use the same measure for calculating - say the distance measure in the case of clustering for all attributes. This is the case with existing algorithms. 

Thanks a lot G. Pavai

In my opinion, there are two ways to handle the problem -- classifier- and attribute oriented.

If classifier-oriented, you use an ensemble classifier that fuses results of multiple subclassifiers. In this case, you can identify a subproblem of your main problem and extract attributes of a particular type, for example, only nominals for building a subclassifier that solves the subproblem. After building multiple subclassifiers, your ensemble classifier can calculate the final result for the main problem.

Alternatively, you use an attribute-oriented approach. I agree with Lewis Maccarter and G. Pavai -- you have to find the normalization form for your data. Probably, it is connected with a data loss, but you don't have to invent the wheel: there exist already some approaches that can help. For example, in WEKA there are attribute filters that handle the problem, for example, NominalToBinary, NominalToString, StringToNominal etc. that convert your attributes from nominals to continuous data.

I provide the links for ensemble classifier and filters in WEKA.

http://weka.sourceforge.net/doc.dev/weka/filters/Filter.html

http://weka.sourceforge.net/doc.packages/ensembleLibrary/weka/classifiers/meta/EnsembleSelection.html

Thank you very much Alexander Osherenko