Even though the classification tree that was built in the previous
section (Figure 4) performs very well for fresh
basalts, this might not necessarily be the case for weathered or
metamorphosed samples. For example, a lot of the power of this tree
depends on Sr, which is considered a mobile element (e.g., Rollinson,
1993). Also MgO, Ni and Rb are species used in Figure
4 that are mobile to some degree. Performance of our
classification, which was based on fresh samples of basalt, cannot be
guaranteed if used for the tectonic discrimination of samples that
have undergone alteration. Therefore, an alternative tree was built
that only uses the so-called high field strength (HFS) cations, which
are characterized by an ionic potential greater than two (Rollinson,
1993), as well as the isotopic ratios of Sr, Nd and Pb, because these
are considered less prone to change during alteration than the
concentrations themselves. The following 28 features were used:
TiO, La, Ce, Pr, Nd, Sm, Gd, Tb, Dy, Ho, Er, Tm, Yb, Lu, Sc, Y,
Zr, Nb, Hf, Ta, Pb, Th, U,
Nd/
Nd,
Sr/
Sr,
Pb/
Pb,
Pb/
Pb and
Pb/
Pb. The resulting pruned tree is shown in Figure
5, while its surrogate splits are given in Table
2. The first two splits, both on TiO
, contribute 85% of the
discriminative power of the tree. The cross-validated
misclassification error of the optimal tree is 16%, i.e., there is
84% chance of a correct classification.