| Logistic regression analysis is used in classification to find out which group an individual belong from a predictor variables set. In classification sometimes we work with data sets with more variables than observations. This is the case of microarray data sets, where there are a relatively small number of observations, generally less than one hundred, and a huge number of features, usually thousands. The following problems in the parameters estimation of logistic regression may occur: over-fitting, unstability and multicollineality. This work explores the logistic regression with Ridge penalty as an alternative to deal with that sort of data sets. It stabilizes the statistical problem, eliminates the numeric degeneracy due to multicollineality and gets low error rates of classification when it is compared with others methods. |
