|
   
Next:Recursive Construction of Models Up:
Model Generation Previous:Identification of Relevant Factors
Black Box Model Construction
Once a set of factors is selected, it is possible establish a linear or nonlinear model that connects the factors to the phenomenon under observation. We assume a total of n observations of the particular
disease state  . For the  observation  , the corresponding observations of the
influencing factors are  . A linear regression model would be expressed as, 
where  ,  is the model coefficient vector, and
The coefficients that minimize the mean square error between the predicted Y
and the collected samples are found by computing, 
The above linear regression models can be extended to include nonlinear terms. A drawback of this approach is the difficulty of associating reasoning, logic, and causality with a single
large model. Consequently, we propose the following mechanism for constructing environmental models. |
