Software

A software package for Bayesian belief network (BN) learning & data
mining is now available for free download at:
http://www.cs.ualberta.ca/~jcheng/bnsoft.htm

The package includes two applications for Windows 95/98/NT/2000:

BN PowerConstructor: An efficient system for learning BN
structures and parameters from data. Constantly update since 1997.

BN PowerPredictor: An extension of BN PowerConstructor for learning
unrestricted BN & Bayes multi-nets based classifiers.

Best regards,

Jie Cheng
Dept. of Computing Science, Univ. of Alberta
email: jcheng@cs.ualberta.ca

############################################################################

Features of BN PowerPredictor:

* Graphical Bayesian network editor for modifying BN classifiers' structure.
* Wrapper algorithm. The system can automatically learn classifiers of
different types and different complexities and choose the best performer.
* Feature subset selection. The system can perform feature subset selection
automatically.
* Two inference modes. The classification can be performed in either batch
mode (data set based) or interactive mode (instance based).
* Supporting domain knowledge.
* Supporting misclassification cost table definition.

Sample Performance of BN PowerPredictor on UCI ML data sets: (Pentium
II-300)

Data set  Cases(Train/Test)  Attr.  Running Time(training) Prediction Accu.

Adult     32561/16281        13      25 Min.                86.33+-0.53
Nursery   8640/4320          8       1 Min.                 97.13+-0.50
Mushroom  5416/2708          22      9  Min.                100
Chess     2130/1066          36      3 Min.                 96.44+-1.11
DNA       2000/1186          60      17 Min.                96.63+-1.03

* The confidence level is 95%. The system performs 300-1000 classifications
per second.