BACKGROUND AND RELATED WORK
The diagnosis of disease can be viewed as a classification
task that sorts subjects into two classes: those with disease
and those without it. Machine learning methods and, in
Manuscript received April 23, 2010. This work was supported in part by
the National Science Foundation Graduate Research Fellowship Program.
I. Tien and S. D. Glaser are with the Center for Information Technology
Research in the Interest of Society (CITRIS), University of California,
Berkeley, CA 94720 USA (emails: itien@berkeley.edu and
glaser@berkeley.edu).
M. J. Aminoff is with the Neurology Department, University of
California, San Francisco, CA 94143 USA (email:
aminoffm@neurology.ucsf.edu).
particular, support vector machines (SVMs) are a powerful
tool for use in classification tasks, especially in high
dimensions. In Parkinson’s disease (PD), SVM has been
used to estimate the severity of tremor, bradykinesia, and
dyskinesia [1]. Studies of gait in PD using SVMs have been
limited to plantar pressure data [2] and ground reaction
forces [3]. The equipment used for these studies only
provides information related directly to the foot hitting the
ground. For example, stride time can be obtained, but no
information can be obtained regarding how the foot moves
during a stride. This limits the understanding of the physical
characteristics of gait that distinguish PD patients from
healthy subjects.