This is arguably a regression dataset — the target attribute is ordinal with three levels: "discharge" (S), "general hospital floor" (A) and "intensive care" (I). However, there are only two records corresponding to the last level, making this not really feasible. Therefore, this was changed into a binary classification task with A and I grouped together.

Attributes "L-CORE", "L-SURF" and "L-BP": {low, mid, high}-> {-1, 0, 1}
Attribute "L-O2", "" and "": {poor, fair, good, excellent}-> {-1, 0, 1, 2}
Attributes "SURF-STBL", "CORE-STBL" and "BP-STBL": {unstable, mod-stable, stable}-> {-1, 0, 1}
