Detecting outliers with the jackknife method
Outlier observations can have a dramatic (often negative) impact on the accuracy of many predictive models. Identifying Outlier observations and handling them appropriately is an important step in the data preparation phase of the CRISP-DM process. Outliers are often defined as observations with extreme values. This is a very limited criterion for defining an outlier. A more robust definition of outlier is an observation that contains a value that is significantly different from what would be predicted by a model built using the other observations in the sample. This definition is more robust as it allows observations to have extreme values if the model predicts an extreme value.
Extreme values can represent normal/predictable outcomes especially in cases of strong variable interaction.
The jackknife method is based on a Monte Carlo simulation where individual data points are held out of the training and testing partition. The overall accuracy of...
