| |
Instructor: Prof.
Madigan
This is a course offered
by statistics department. As we know data mining requires very
strong statistics background. Then I am trying to make some progress
in this area. As a special topic here, we are discussing the data
mining both from statistics perspective and from computer science
perspective. But we will have to talk more on predictive models
and pattern, score functions for data mining algorithms, exploring
data, data analysis, and so on. We are required to finish a big
project individually, plus one proposal by spring break and a
final presentation. So a big job!
As for project, I've focused
on the big Bayesian Network. What I implemented is an ALARM (A
Logical Alarm Reduction Mechanism) belief network. It's a medical
diagnostic alarm message system for patient monitoring. Given
a version of ALARM network, first to simulate a dataset with 10,000
cases using Monte Carlo simulation method. Second, to use the
K2 algorithm to reconstruct (learn) the most possible belief network
structure. Finally based on the newly created network structure,
predict 8 diagnostic problems respectively given all other measurements,
and output the percentage of correct prediction for each problem.
So when ALARM is given patients' measurements, it outputs a probability
distribution over a kind of possible problem.
You can download my proposal
and presentation slides here. Also my final project report is
available upon request.
References:
- [Cooper and Herskovits,
1992] Cooper, G. and Herskovits, E. (1992). A Bayesian method
for the induction of probabilistic networks from data. Machine
Learning, 9:309-347.
- [David Heckerman] David
Heckerman (1995). A tutorial on learning with Bayesian networks.
MSR-TR-95-06.
- Beinlich, I.A., Suermondt,
H.J., Chavez, R.M., & Cooper, G.F. (1989). The ALARM monitoring
system: A case study with two probabilistic inference techniques
for belief networks. Proceedings of the Second European Conference
on Artificial Intelligence in Medicine (pp. 247-256). London,
England.
- http://www.norsys.com/networklibrary.html.
The true ALARM network structure can be found here.
- http://www.cs.ualberta.ca/~jcheng/bnpchlp/.
|