Converting Classification Runs to Logistic Regression

GeneXproTools allows you to convert runs created within the Classification Framework to Logistic Regression. This means that you’ll be able not only to access the raw scores of your classification models (that is, the model scores before they were rounded to either 0 or 1) but also to use them to assign probabilities to your categorical classifications in the Logistic Regression Framework.

Any old Classification run can be converted to Logistic Regression. But you may also consider creating new ones with the sole purpose of exploring all the Classification fitness functions (there are a total of 15 different fitness functions in the Classification Framework, therefore a nice addition to the fitness functions of the Logistic Regression Platform). Then you can see how good these models perform in the Logistic Regression Framework by analyzing their ROC Curves and Optimal Cutoff Points, Quantile Tables and Gains and Lift Charts, and also their Logistic Fit and Confusion Matrices. And if they are not yet what you need, you can always use them as seed (either here in the Logistic Regression Platform or back in the Classification Platform) to create better models with them. You can obviously repeat this process for as long as you wish, until you obtain the right model for your data.

To convert a Classification run to Logistic Regression you need to:

1. Within the Classification Framework, choose Convert To Logistic Regression in the Logistic Regression menu.
   This opens the Save As dialog box and also asks if you want to save the current run before converting it to Logistic Regression. This way you will be able to come back to it if you need to.
2. Type the run name for the new Logistic Regression run and then click Save.
   When you click Save, GeneXproTools takes you immediately to the Logistic Regression Framework. Note that some of the statistics of the models in the Run History are updated, showing now the R-square instead of the Classification Accuracy. Note, however, that these R-square values were evaluated using rounded model scores (0 or 1) in the Classification Framework and will, therefore, differ from the ones evaluated here using the raw model scores. By choosing Refresh All you can rapidly update these values to their true values in this new context. The default fitness function for the new run is the Correlation Coefficient as this fitness function, together with the similar R-square fitness function, produces the best results with the standard 0/1 class encoding. But if you are using a different encoding, you may want to experiment with other fitness functions.