Linear Regression Diagnostics

This component provides some important linear regression diagnostics which are missing in standard KNIME nodes.

Use the dialog to select the target variable of the linear regression model used.

Attention: The component does not scale well for large datasets (More than 5,6 k rows).
Application on large datasets could be take a lot of time.

Connecting this node to the "Linear Regression Predictor" allows to create a dashboard useful to:
- Identify influencial rows in OLS estimates using Cook's distance;
- Inspect the distribution of errors (it should be approximately normal);
- Visualize the scatterplot of the predicted values vs the real values;
- Visualize the scatterplot of the errors vs the predicted values;
- Visualize the scatterplot of the errors vs true values;
- View the ANOVA TABLE and the F statistic with relative p-value;
- Check the residuals standard error.

Input ports

1. Type: PMML
   PMML Linear Regression
   Connect the estimated Regression Model by using the PMML Port
2. Type: Table
   Coefficients table
   Connect to "Coefficients and Statistics" data output of the Linear Regression Learner node
3. Type: Table
   Linear Regression input table
   Connect to the same table which feeds the Linear Regression Learner node

Output ports

1. Type: Table
   Cook's Distance
   Cooks distance for each row the dataset