Page 314 - Machine Learning for Subsurface Characterization

Classification of sonic wave Chapter 9 275

TABLE 9.3 Classification report of the voting classifier on the testing dataset
for the classification-based noninvasive characterization of material
containing static discontinuities of various dispersions around the primary
orientation.
Material type Precision Recall F 1 score Support
0 0.73 0.78 0.75 3003
1 0.53 0.4 0.46 2999
2 0.66 0.78 0.72 2998
Avg/total 0.64 0.65 0.64 9000

5.2 Characterization of material containing static discontinuities
of various primary orientations
5.2.1 Background
In this section, nine classifiers (discussed in Section 4.1) process compressional
wavefront travel times to categorize materials containing discontinuities in terms
of the primary orientations of the discontinuities. The von Mises distribution is
used to generate 100 discontinuities of specific dispersion around various primary
orientations. The presence of dispersion leads to randomness in the orientations of
the discontinuities. Various networks of discontinuities with distinct primary ori-
entation and dispersion are created as listed in Table 9.4. The four datasets with
associated user-assigned labels, as listed in Table 9.4, are processed by the nine
classifiers to learn to noninvasively characterize the material containing disconti-
nuities in terms of the primary orientation in the presence of specific dispersions in
the orientation. User-assigned labels denote various material types.
For the Datasets #1 and #2 (Table 9.4), user assigns one of the four labels,
denoting four material types, to each sample because the samples are generated
for materials containing discontinuities having primary orientation of 0, 45, 90,
or 135 degrees and concentrations of 10 and 50, respectively, in terms of kappa.
A concentration of 10 indicates dispersion of +50 to 50 degrees around the
primary orientation. A concentration of 50 indicates dispersion of +20 to
20 degrees around the primary orientation. Similarly, for the Datasets #3
and #4 (Table 9.4), user assigns one of the eight labels to each sample because
the samples are generated for materials containing discontinuities having pri-
mary orientation of 0, 22.5, 45, 67.5, 90, 112.5, 135, and 157.5 degrees and con-
centrations of 10 and 50, respectively, in terms of kappa. Dataset #2 and Dataset
#4 are generated by materials containing discontinuities of lower dispersion
around primary orientation as compared with Dataset #1 and Dataset #3, respec-
tively; in other words, the materials that generate Dataset #2 and Dataset #4
have more directionally aligned discontinuities. In Fig. 9.23, the upper and