leaf nodes perform a numeric prediction
using a linear or nonlinear regression
over the target values associated with the leaf node

Each leaf node represents a model,
as a function of input variables,
where the model minimizes a MSE or SDR for the subset associated with the leaf node
Induced in the same way as a regression tree
Example
Time-series prediction

All regression tree aspects also apply to model trees:

predicts a numerical value (not a discrete class value)
same workflow as induction of classification trees
- Algorithm
Prediction math
Measure of inhomogeneity (not entropy)
- MSE
- SDR

Induced until MSE or SDR is small enough:
- likely to overfit, so post-pruning is also applied
Large datasets: windowing
Data types: same numeric approach
Same process is applied to extract rules from a regression tree
Data quality issues