This node performs Target Shuffling by randomly permuting the values in one column of the input table. This will break any connection between input variables (learning columns) and response variable (target column) while retaining the overall distribution of the target variable. Target shuffling is used to estimate the baseline performance of a predictive model. It's expected that the quality of a model (accuracy, area under the curve, R², ...) will decrease drastically if the target values were shuffled as any relationship between input and target was removed.
It's advisable to repeat this process (target shuffling + model building + model evaluation) many times and record the bogus result in order to receive good estimates on how well the real model performs in comparison to randomized data.
Target shuffling is sometimes called randomization test or y-scrambling. For more information see also Handbook of Statistical Analysis and Data Mining Applications by Gary Miner, Robert Nisbet, John Elder IV.
- Type: TableInput dataAny data table