1 min read

Predictive modeling

Predictive modeling

Lesson from radioimmunology prediction modeling.

First look at the data.

  • The predictive variable names must be identical in both train and test sets.
  • Output variable should be well defined. Unvalenced?
  • Look at the missing values. Consider imputation

Preprocessing

  • Resolving skewness
    • Box-Cox transformation should be done in non-negative varialbes.
    • Use Yeo-Johnson transformation for variables including negative value.

Errors when prediction of test set

  • “pred method can not take rf object” like error means random forest (rf) modeling fails.

Feature selection vs Penalized model

  • Use Penalized model.
  • No more stepwise feature selection.