I am working with the following problem: My X data are far-field measurements in a surface (90k rows with 20 features) where the features are e.g. column 1 = measurement at 0 degrees, column 2 = measurement at 10 degrees, etc. The X features are in general correlated. Each row corresponds to a different sample. My target y values are 90k rows and 6 columns, which are positions in polar coordinates of the sources producing the surface data. My problem is a multioutput regression problem. My data are stored in .csv and loaded in the NN models.