We are lucky here that the mean (black dot) fall on a land area, but only 8% of the testing set enclosed in the shade region is water (negative predictive value): a pretty bad approximation if you were a parachutist expecting to fall on land! Worse, the map metaphor relies on only two dimensions. As you add dimensions (the tree in figure \ref{273738} has six balances), the proportion of the healthy hypershape in the hypercube will decrease considerably \cite{Nowaki_2017}. Moreover, selecting an alternative balance scheme would rotate the axes and greatly affect the range-based diagnosis. Although results conform with the literature, they are biased and likely inaccurate.
Using ellipsoids
In a paper on mango ionomics \cite{Parent_2013a}, I analyzed the ionome with an in-house classification technique based on the Mahalanobis distance to classify specimens between high and low crop yields. The algorithm iterated to isolate true high yielders (high yields correctly diagnosed as high yields) as reference. Let’s see if the Mahalanobis distance is useful (shaded region in figure \ref{385754}).