Lab 8: Bayesian Networks

Introduction

In this lab we will be working with a Bayes' net tool developed at the University of British Columbia's AIspace project. The tool can be found by visiting the Belief and Decision Networks page and clicking the link labeled "Click here".

Exercises

• Disease Diagnosis

  The goal of this exercise is to design a Bayes' net to aid in the diagnosis of some small set of diseases. You should select at least three diseases, as well as at least three possible symptoms. You should then create edges from each disease to the symptoms that that disease could potentially cause. (Remember that arrows go in the direction of causation: diseases cause symptoms, so arrows go from diseases to symptoms.) Finally, you should fill in the probability tables for each node with reasonable values. If you are interested in tackling some other inference task, feel free. For example, you could diagnose automobile problems rather than human diseases. Once you have completed your network:
  
  
  • Assign three different sets of symptoms to your patient, and query the disease nodes to see which disease is most probable in each case. Write down the results. Do the networks diagnoses conform to your expectations?
  • After indicating a set of symptoms, try excluding one possible disease by setting its observation to F. Does this change the probability assigned to the other diseases?
  • Augment your network with information about behaviors that might have caused the diseases that you are diagnosing. For example, since smoking can cause lung cancer, you might add a "Smokes" node with a link to a "Lung Cancer" Node. Assign observations to those behaviors and analyze the effect that this has on the probabilities assigned to both diseases and symptoms. Then try working in the other direction: try assigning symptoms to see what effect this has on the probabilities assigned to behaviors. You don't need to write anything down for this question.
  • Draw or print your final network. Write down the probability tables associated with one of the nodes at each level: one activity node, one disease node, and one symptom node.