Sometimes it is useful to know how much our belief in a particular node is influenced by findings at other nodes. We want to know how sensitive is our belief in this node's value to the findings of other nodes. If it is very sensitive, we may want and it is important for us to know the state of that node, then we may want to invest the effort in determining the values of all the nodes that substantially influence it. Netica can compute a node's sensitivity to findings easily.

In this tutorial we will look at the net ALARM. 'ALARM' stands for 'A Logical Alarm Reduction Mechanism'. It is a medical diagnostic alarm message system for patient monitoring and is described in BeinlichSCC89. It is a real-world application of Bayes nets. It is too complex to describe here, but it will suffice for our purposes. You will likely want to bring it up in Netica to work with, for better visibility.

Let us suppose we want to know which nodes can most influence our knowledge of "Heart Rate". Click on the top of the title region of the "Heart Rate" box to select it, and then choose Network->Sensitivity to Findings.

A detailed report will be produced in the "Netica - Messages" window (if you can't find it, click on Window->Messages to raise it). Here is what the report looks like:

Probability of new evidence = 100 %, of all evidence = 100 %.
-------------------------------------------------------------------------

Sensitivity of 'HR' to findings at 'HR':

  Probability ranges:   Min         Current     Max         |  RMS Change
  Low                   0           0.04678     1           |  0.2112    
  Normal                0           0.4169      1           |  0.493     
  High                  0           0.5363      1           |  0.4987    

  Quadratic scoring  = 0.2996

  Entropy reduction  = 1.215  (100 %)
-------------------------------------------------------------------------

Sensitivity of 'HR' to findings at 'HRBP':

  Probability ranges:   Min         Current     Max         |  RMS Change
  Low                   0.0009137   0.04678     0.6431      |  0.1652    
  Normal                0.008143    0.4169      0.9609      |  0.4617    
  High                  0.03796     0.5363      0.9909      |  0.4671    

  Quadratic scoring  = 0.2467

  Entropy reduction  = 0.9799  (80.7 %)
-------------------------------------------------------------------------

-------------------------------------------------------------------------

Sensitivity of 'HR' to findings at 'PAP':

  Probability ranges:   Min         Current     Max         |  RMS Change
  Low                   0.04307     0.04678     0.04701     |  0.0009153 
  Normal                0.384       0.4169      0.419       |  0.008136  
  High                  0.534       0.5363      0.573       |  0.009052  

  Quadratic scoring  = 7.235e-005

  Entropy reduction  = 0.0002387  (0.0196 %)
-------------------------------------------------------------------------

Sensitivity of 'HR' due to a finding at another node:

Node                Mutual      Quadratic   
----                Info        Score       
HR                  1.21495     0.2995618   
HRBP                0.97993     0.2466639   
HREKG               0.81844     0.2124077   
HRSat               0.81844     0.2124077   
Catechol            0.68583     0.1587396   
CO                  0.52149     0.1401417   
SaO2                0.16188     0.0461771   
PVSat               0.12620     0.0362042   
VentAlv             0.11472     0.0330883   
ArtCO2              0.11419     0.0333368   
ExpCO2              0.10006     0.0295365   
InsuffAnesth        0.09786     0.0283655   
BP                  0.09734     0.0268312   
TPR                 0.09482     0.0282133   
VentLung            0.07704     0.0224051   
MinVol              0.06422     0.0189986   
Press               0.04806     0.0143849   
Shunt               0.04013     0.0117534   
VentTube            0.03752     0.0109797   
Intubation          0.03224     0.0093943   
Disconnect          0.01694     0.0049907   
KinkedTube          0.01427     0.0041830   
VentMach            0.01091     0.0032041   
MinVolSet           0.00360     0.0010587   
FiO2                0.00304     0.0009002   
PulmEmbolus         0.00254     0.0007571   
Anaphylaxis         0.00208     0.0006266   
PAP                 0.00024     0.0000723   
LVEDVolume          0.00000     0.0000000   
StrokeVolume        0.00000     0.0000000   
LVFailure           0.00000     0.0000000   
Hypovolemia         0.00000     0.0000000   
ErrLowOutput        0.00000     0.0000000   
CVP                 0.00000     0.0000000   
PCWP                0.00000     0.0000000   
ErrCauter           0.00000     0.0000000   
History             0.00000     0.0000000  

The exact meaning of everything in the report is described in Netica's onscreen help.

At the end of the report is a list of all the nodes, with the ones whose findings are most likely to produce the greatest change in the belief of Heart Rate listed first. The first column of numbers provides a numerical degree.

Note, that not unexpectedly, the most influential nodes of "Heart Rate", are its parents and children.

Notes on Sensitivity to Findings

• The report takes findings currently entered into account.
• The nodes are referred to by name, rather than by title. This is because names are size-limited and thus easier to format. You can display nodes with their names rather than titles by choosing Style->Name.

... TBD ...