Introduction to Bayesian belief networks

  1. Get familiar with the basic operations and principles of building the Bayesian Belief Networks and Influence Diagrams by refreshing the lecture material and/or reading some tutorial. Practice using the demo tool available from the AISpace repository (see References below).

  2. Get familiar with some real practice dataset, such as the Bike Sharing dataset (see References). Read the description, download the set.

    It is possible to use a different dataset for this assignment, as long as it is a true statistic of some real phenomenon, and it has enough data to allow computing various conditional probabilities. For the purpose of this assignment, the network should be limited to approximately 5-7 nodes. Several hundred data records is the minimum dataset size.

  3. Experiment with the data using a Bayesian Network building tool, which has the network learning capability, such as the B-course at the University of Helsinki (See References). Programs such as Genie, Hugin, and Netica, also have this capablity. Experiment building the network using different subsets of the data attributes.

  4. For the selected random variables and the network graph created partially by the automated learning process and partially by manual corrections, build your own Bayesian Belief network using the AISpace tool.

    Experiment with several different queries (obtaining the probabilities of some random variable) with different combinations of knowledge (observed random variables). Select the most interesting result and put it in the report.

  5. Think about some real decision, or action, that could be taken using the knowledge provided by such network. Add it as a decision node to the network. Then define your own utility distribution in the form of a single Utility node.

    Experiment with making decisions using your Decision Network. Put an example decision in the report from the assignment.

References:

http://aispace.org/bayes/
http://www.aispace.org/bayes/version5.1.10signed/bayes.jar
https://archive.ics.uci.edu/ml/datasets/Bike+Sharing+Dataset
http://b-course.cs.helsinki.fi/obc/depend.html
http://www.hugin.com/
http://www.bayesfusion.com/downloads-rxu9q
http://www.norsys.com/netica.html