Subject: | |
From: | |
Reply To: | |
Date: | Tue, 30 Nov 1999 10:00:19 -0500 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Dear Colleagues,
We continue our Colloquium.
When: Thursday, December 02, 2:30 p.m.
Where: Metro 207
Speaker: Akintoye Oloko, graduate student, College of Engineering, UTC
Classification of a Seven Feature Data Set
ABSTRACT: Two sets of data from two sensors are given. One sensor has four
features, the other have three. The first set of data will be called the
training set. It contains a classification feature that tells which class a
feature belongs to. The data is classified into four classes were the first
class is considered dangerous and the other three safe. This problem can be
treated as a two-class problem, but for academic reasons the problem will
be treated as a four-class problem.
Assuming a Gaussian distribution a probability density function is
generated for each class and then summed to generate a probability function
for our entire sample space. This can be done if we assume that our classes
are disjoint and cover the entire sample space. Using Bayesian decision
theory the {\it a posterior} probability of the class given the pattern is
then calculated. This is then used to generate a loss function for each
class. The class with the minimum loss function is then chosen as the class
the feature belongs to.
Accepting a decision gives rise to a classification error or correct
decision both on the class and the entire space. These values are then used
to see how well the features classify the data.
Comment. THIS IS OUR LAST COLLOQUIUM IN 1999.
Colloquium Committee
Boris Belinskiy
Department of Mathematics
University of Tennessee at Chattanooga
615 McCallie Ave
Chattanooga, TN 37403-2598
Ph. 423-755-4748
Fax 423-755-4586
e-mail [log in to unmask]
|
|
|