The cost matrix is an important component of most fitness functions for
logistic regression and
logic synthesis. It allows you to attribute a cost to the
four different classification outcomes, commonly shown in the
true positives (TP), true negatives (TN),
false positives (FP) and false negatives (FN). The cost associated with
correct classifications is positive and negative for misclassifications.
The default values GeneXproTools provides are a good neutral starting point,
as they are based on the proportion of positive and negative records in the training dataset.
GeneXproTools evaluates the overall weight of the cost matrix and combines it
with other fitness components, such as the
area under the ROC curve or the
The weight of the cost/gain balance was fine-tuned for each fitness function
in order to ensure the best possible performance on a wide range of datasets.