An approach to classifying four-part music in multidimensional space - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 26

6.5 Classifying from Among Two Composers
For all three evaluation techniques, the averages of each trial, when classifying among two
composers, came out well above 50%—the value expected from a two-composer coin toss. In
fact, no individual trial fell below 50%.
Bach Versus Dykes—All Metrics

Technique Correctness (%)

Unweighted Points 82.5
Weighted Points 86.8
Euclidean Distance 71.5

Bach Versus Smart—All Metrics

Technique Correctness (%)

Unweighted Points 92.1
Weighted Points 89.3
Euclidean Distance 69.3

Dykes Versus Smart—All Metrics

Technique Correctness (%)

Unweighted Points 74
Weighted Points 82.9
Euclidean Distance 69.3

The best technique overall was Weighted Points, demonstrating the strongest performance
in two out of the three classifications.

6.6 Classifying from Among Three Composers
For all three evaluation techniques, the averages of each trial, when classifying among three
composers, came out well above 33.3%—the value expected from random, three-way guess-
ing. In fact, no individual trial dipped below 33.3%. The technique that worked best was Un-
weighted Points followed by Weighted Points at a close second.