Prof. H. Is.
ıl Bozma
Bogazici University, Electric Electronic Eng., 20.11.2024
Due date: 28.11.2024 (19:00)
EE 573 Pattern Recognition - Project 3
Consider the given dataset and select 5 classes. For each class c = 1, . . . , 5, let Dc be the learning set and
Tc denote the test set.
Suppose each class c is associated with two probability density functions as defined by Gaussian N (µ+ +
c , σc )
− − + + − −
and N (µc , σc ). Use MLE to find µ̂c , σ̂c and µ̂c , σ̂c .
Prepare a table containing all the estimated parameter values for all the classes.
Use the resulting MLE values to assign sample in the test set to a class. What is P (error). Compute
the resulting recall and precision rates for each class.
Now assume that the positive set for each class is defined as a mixture of K Gaussians - namely p(x | ωc ) =
PK
m=1 γcm N (µcm , σcm ) where γcm , µcm , σc,m ,m = 1, . . . , K are unknown parameters. Apply expectation
maximization to determine the unknown parameters - using the learning set. Consider three alternative K
values. For each K value,
Prepare a table containing all the estimated parameter values for all the classes.
Assign each sample in the test set to a class. What is P (error). Compute the resulting recall and
precision rates for each class.
Based on your results, which K leads to best performance. Explain your reasoning.
Pls hand in a short report that gives the decision rules for each part, your results along with their comparison
as required along with your code and a readme file that explains how to run your code.
