Please enable JavaScript.
Coggle requires JavaScript to display documents.
Regression Summary - Coggle Diagram
Regression Summary
Regression Tree
Split Rule: splits in class are determined by SSD, lower SSD means more alike
Output: To predict the outcome for new attributes, they are tested against the regression tree
Leaf nodes: separated into class, average of values in the class
Pros/Cons
Pros: No parametric assumptions, good variable selection, robust to outliers
-
Regression Input/Output
Input: Observation, attributes (dependent, independent)
-
-
KNN
Steps:
- Normalize data if needed (only attributes needed to compute distance, not Y)
- For every new observation, pick nearest K observation
- Pick distance metric and choose K, picked to minimize RMSE
- Predict new observation by taking average of nearest neighbor observation
Pros/Cons
Pros: simple, doesn't make assumptions about data distributions
Cons: If you have a lot of columns, you have to have a lot of rows, very slow
Evaluating Regression
-
Metrics: Mean absolute error, Root mean squared error
-
-