I have a dataset with a total of 103 cases. There are three distinct values for my dependent variable each of which having different size, i.e. 43, 42, and 18. I am using SPSS to classify my data using a Decision Tree classifier. My aim is to create spilling rules whereby I can predict the future events. However, I am not sure about the values I should input for minimum number of cases for each of the parent and child nodes. Decreasing the respective values lead to an increase in the number of terminal nodes and the depth level and of course the accuracy of classification, which is a desired aim for me. Anyhow, I am not sure as to whether there is a limit, kind of threshold that lowering the values below that makes the result meaningless.
More than the size of the data set what influence the most in the tree size, is the heterogeneity of the data.
Let’s suppose that the data set is so homogeneous, that with only two rules we can handle it, in this case if you take more point that honor the same rules the decision tree will be the same. In the other hand with a smaller data set but with high heterogeneity when we increase the size of the data set new rules could be necessary to explain those new cases.
One idea to try to understand the numbers of rules on the decision tree could be to run a k-means whit different values of k in order to obtain the number of cluster we have in our data, this number of cluster could be a indicator of the tree size.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS