It should be mostly based on the use case. You decide which algorithms you need mostly based not on the data (though data might pose some restrictions) but on what you want to do with the data.

Algorithms in general - and you can consider some of the most famous ones like Sort, RSA, Data compression (huffman code), etc... are irrespective of any data context - binary is neutral.

I guess for data scientists the premise somehow holds. For example if we need to group data for its similarity (clustering) from distances between 'points' what matters most is such idea's adherence to the real case, and eventually not whether we are talking about similitudes in text mining or geographical locations for product delivery planning. So I guess the 'units' of the data probably are not that important, unless they contain relevant information about the use case. 

But probably the type of data will also have to be taken into account, for example, its format or precision might not be compatible with some algorithms.

Good luck.

Thanks much Fernando for your detailed reply. It helps a lot. 

I assume that algorithms obey the use case (and probably a telecom application has different requirements than a retail one). But, even in the same type of application, as the data source is important, in particular for data science, you probably need to add a preprocessing stage that transform the original data to a format suitable with your algorithm.

Thanks Rodrigo. Yeah, am considering quite a few pre-processing steps before I go to selection stage of an algorithm.