Specify the default methods for classification trees. The changes you make to the defaults remain until you change them again, even after you exit Minitab.
In This Topic
Choose the splitting method to generate your decision tree. You can compare the results from several splitting methods to determine the best choice for your application.
Gini: The Gini method is the default method. The Gini method works well across many applications. The Gini method usually generates trees that include small nodes with a high concentration of the response of interest.
Entropy: The Entropy method is proportional to the maximum of certain likelihood functions for the node.
Twoing: The Twoing method is only available with a multinomial response. The Twoing method usually generates more balanced splits than the Gini or Entropy methods. For a binary response, the Twoing method is the same as the Gini method.
probability: The probability tree tends to be larger than the Gini tree. Use the probability method when you are interested in the performance of a few top nodes.
Criterion for selecting optimal tree
Choose between the following criteria to select the tree in the results. You can compare the results from different trees to determine the best choice for your application.
misclassification cost: Select this option to display results for the tree that minimizes the misclassification cost.
Within K standard
errors of minimum misclassification cost; K=: Select this option to display results for the smallest tree with a misclassification cost within K standard errors of the minimum misclassification cost. By default, K=1, so the results are for the smallest tree with a misclassification cost within 1 standard error of the tree with the minimum misclassification cost.
number of cases to split an internal node
Enter a value to represent the minimum number of cases an internal node to be split. The default is 10. With larger sample sizes, you may want to increase this minimum. For example, if an internal node has 10 or more cases, Minitab will try to perform a split. If the internal node has 9 cases or less, Minitab will not try to perform a split.
The internal node limit must be at least twice the terminal node limit, but larger ratios are better. Internal node limits of at least 3 times terminal node limits allow a reasonable number of splitters.
The default value is 10.
number of cases allowed for a terminal node
Enter a value to represent the minimum number of cases that may be separated into a terminal node. The default is 3. With larger sample sizes, you may want to increase this minimum. For example, if a split would create a node with less than 3 cases, Minitab will not perform a split.
The default value is 3.
Missing value penalty
Enter a penalty value for a predictor with missing values. Because it is easier to be a good splitter with less data, predictors with missing data have an advantage over predictors without missing data. Use this option to penalize predictors with missing data.
0.0 ≤ K ≤ 2.0, for example:
K = 0: Specifies no penalty.
K = 2: Specifies highest penalty.
High level category penalty
Enter a penalty value for categorical predictors that have many values. Because categorical predictors with many levels can distort a tree due to their increased splitting power, they have an advantage over predictors with less levels. Use this option to penalize predictors with many levels.
0.0 ≤ K ≤ 5.0, for example:
K = 0: Specifies no penalty.
K = 5: Specifies highest penalty.
Display of graphs and tables
Rates from confusion matrix
Choose the rates that you want to display on your confusion matrix.
True positive: True positive rate (TPR) - the probability that an event case is predicted correctly.
False positive: False positive rate (FPR) - the probability that a nonevent case is predicted incorrectly.
False negative: False negative rate (FNR) - the probability that an event case is predicted incorrectly.
True negative: True negative rate (TNR) - the probability that a nonevent case is predicted correctly.
characteristic (ROC) curve
The receiver operating characteristic (ROC) curve shows the ability of a tree to distinguish between classes. The ROC curve plots the true positive rate (TPR) against the false positive rate (FPR).
The cumulative gain chart illustrates the effectiveness of the model in a portion of the population. The gain chart plots % class versus % population.
The lift chart illustrates the effectiveness of the predictive model. The lift chart plots cumulative lift versus % population and displays the difference between results obtained with and without the predictive model. You can specify Cumulative or Not
cumulative for this chart.
Choose whether to display the best nodes, the worst nodes, or both for the Effectiveness of classification table and the Criteria for classifying subjects table.
Best: By default,
Minitab displays the best terminal nodes. The best nodes have the highest event probability (binary) or highest class probability (multinomial) values. For a binary response, the best nodes have event probabilities close to the two end cases of 1 or 0.
Worst: Select to display the worst terminal nodes. The worst nodes have the lowest event probability (binary) or lowest class probability (multinomial) values. For a binary response, the worst nodes have event probabilities close to the middle value of 0.5.
Worst: Select to display the best and the worst terminal nodes.