|
  
Next:Black Box Model Construction Up:Model Generation Previous:Model Generation
Identification of Relevant Factors
There exist well established statistical techniques to select factors that are correlated to particular phenomenon. In particular, the evaluation of the covariance between two variables,  and  , is commonly used to measure the correlation between a factor and the phenomenon under investigation: 
A covariance other than zero indicates that the two factors are either positively (if the covariance is
greater than zero) or negatively (if the covariance is less than zero) correlated. In the HPS model, might refer to the amount of precipitation in a given month, while refers to the population growth of mice. Other possible predictor variables include precipitation, biomass (often
measured in terms of vegetation index), existence of wet habitat, and density of susceptible population broken down by age and sex. The goal is to select a set of nonredundant or minimally redundant
factors that serve as good indicators or predictors for the disease. The common procedure for selecting relevant factors is to select
those factors that are strongly correlated to the phenomenon being investigated (e.g.,  ) and
are uncorrelated (or weakly correlated) with other factors. |
