Decision Support Systems (DSSs) are devoted to the management of large volumes of data. Its main goal is to simulate decision-making procedures that actually simulate (or show) some intelligence. This is why the wikipedia see DSSs well related to business processes ---as they are, indeed.
Machine Learning (ML) is used to infer or derive new knowledge from existing one usually expressed in the form of static data and optionally, as dynamic data or rules. The new knowledge is also expressed in one or both forms.
In general, DSSs and ML are different in their scope. ML is not expected to serve to take decisions though some mechanisms produce new data in such a form that a recommendation becomes apparent (e.g., when using clustering algorithms). Instead, ML can serve in the back for DSSs which I see as a more general system: it is not only in charge of taking decisions but also of managing data, the human-machine interaction (e.g., implementing Mixed-Initiative procedures), and a number of additional modules such as an explanatory module which provides additional evidence of the decisions taken.
None of these processes (necessarily) exist in ML: no need to manage data in any particular way (and indeed data is usually assumed to come either in a declarative way, e.g., PROLOG, or in a large Database conveniently designed, e.g., as in RapidMiner); no particular interface is usually required since the interesting product is the new knowledge which is writen usually in the same format than the input; no need to produce explanations of the learnt knowledge (though, admittedly, some mechanisms produce a trace that can be revisited to verify the learnt knowledge).
Other than these rather theoretical (but not really) considerations, let me emphasize that many (if not all) DSSs usually consist of Expert Systems (even nowadays) while ESs are not a form of ML.
Hope this helps,