|
MARS (Multivariate Adaptive
Regression Splines) was used to automatically
identify the most relevant parameters for a
classification problem. In the present case the
algorithm yields that out of 68 parameters only
two are needed to separate the sample data into
“deciduous trees“ and “coniferous trees“.
Compared to the number of input parameters the
sample size was very small (18), which shows
that the algorithm can deal with rather low
signal to noise ratios. The computational time
(Pentium III, 900 MHz) was less than a second in
the present case.
 |
Download
article here