Use to build a partition-based model (Decision Tree) that identifies the most important factors that predict a categorical outcome (classify) and use the resulting tree to make predictions for new observations.
Classification Trees
- From an open JMP® table, select Analyze > Predictive Modeling > Partition.
- Select a nominal or ordinal response variable from Select Columns and click Y, Response.
- Select explanatory variables and click X, Factor.
- If desired, enter the Validation Portion or select a validation column and click Validation (JMP Pro only). A randomly selected validation set of 30% was used in this illustration. Thus 70% for training.
- In JMP Pro only, select the tree Method: Decision Tree (the default), Bootstrap Forest, Boosted Tree, K Nearest Neighbors or Naive Bayes. JMP Pro also allows you to specify a validation column.
- Click OK.
JMP initially displays a graph showing the proportion of observations in each response level. - Click the Split button. The observations will be split into two nodes, or leaves. The graph will update to reflect the split and a tree diagram describing the split in more detail will be created.
Note: Click on the top red triangle and select Display Options > Show Split Counts to show Counts, Rates (proportion of observations) and Probs (predicted probabilities) in each leaf.
- Click Split to continue additional splits. Click Prune to remove a split.
If a validation set is used, click Go to perform automatic splitting and pruning optimizing the fit on the validation data. Here only two splits were performed.
Interpretation for the first two splits (Response is Claim (Y/N)):
• There are 1,179 obs in the left leaf, corresponding to AgeClass(Young). 768 of those (65.1%) are Claim(Y/N) = N and 411 (34.8%) are Claim(Y/N) = Y.
• There are 12,758 obs in the right leaf, corresponding to AgeClass(Elder). The response rate is 90.2% for N and 9.8% for Y in that node.
• For the 12,758 observations in the AgeClass(Elder), the second split, based on the Rating Class variable, is at Rating Class (D,C) and Rating Class (B, A).
Notes:
For additional options, such as Column Contributions, ROC and Lift Curves, click the top red triangle. Other options, such as Save Prediction Formula are available from the top red triangle > Save Columns. Select Decision Threshold under the top red triangle to display correct and incorrect classification rates for the model including the ability to evaluate those rates under different cutoff values. The default is 50%.
Visit Predictive and Specialized Models > Partition Models in JMP Help to learn more.