Multivariate analysis focuses on the results of observations of many different variables for a number of objects. In this sense most machine learning methods lie within this group; I cannot imagine performing regression, without having more than one statistical outcome variable at a time. However, what about methods such as clustering and anomaly Detection? Although we could easily extend the concepts to a high-dimensional space, and perform a multi-dimensional clustering (for instance SOM), I believe most clustering and anomaly applications work for a single-variable. In that particular case, could they be consider univariate methods?
[link][2 comments]