The Power of Random - Course: Using Perturbation Experiments to Improve Model Accuracy and Interpretation, Austin


KNIME is hosting the following one-day course during the KNIME Fall Summit in Austin on November 7, 2018:

The Power of Random: Using Perturbation Experiments to Improve Model Accuracy and Interpretation

Predictive modelers often start learning how to build models with linear methods and statistical models. These approaches usually assume smaller data, known distributions, no missing values, and more, and as a result, make building and assessing the models straightforward, blessed with many very good metrics to use to judge model accuracy and the influence of individual predictor variables. As we build models using non-parametric, highly nonlinear techniques, many of these measures no longer make sense or are impossible to apply.

This workshop summarizes a half dozen ways to use randomization in the model building process. For each, principles describing the approach will be provided, followed by demonstrations of the techniques using KNIME.

This course is intended for Data Scientists, Statisticians, Mathematicians, Computer Scientists, and IT Professionals who build and interpret predictive models. Prior predictive modeling experience is very helpful.

Register Now

Course Content

  • Randomization principles for Sampling
  • Randomization for Missing Value Imputation
  • Randomization in building models (Random Forests)
  • Randomization to Assess Confidence in Model Accuracy: Target Shuffling
  • Randomization to Assess Variable Importance: Input Shuffling
  • Randomization to Create Model Prediction Confidence Intervals

Download preliminary agenda.

Other Courses