Diversified
Multiple Decision Trees
Introduction
This program uses diversified multiple decision trees to make high accurate predictions on high dimensional data sets. See explanations and experimental results in A Maximally Diversified Multiple Decision Tree Algorithm for Microarray Data Classification.
This software tool works on Windows. (embedded in Weka.)
Get program
Run program
Click run.bat to run the program.
Click Explorer at the GUI interface of Weka
Open amlall.arff file
Choose Classify -> Choose -> meta -> MDMT_J48
Set parameters as the following. You may click More to read explanations about options.
Click start to run and should get the following results
Other options
eliminateAllAttributes: TRUE removes
all attributes used by the previous tree, FALSE removes only the root node of
the previous tree.
votingType: 0 unweighed vote of
trees. 1 weighted by the confidence of the decision leaf. 2 weighted by the support
of the decision leaf, 3 weighted by the Laplace accuracy of the decision leaf.
useDiversity = TRUE
A new tree has to satisfy the
diversity requirement specified by minDiversity with all existing trees used by
MDMT. Diversity is defined as the following.
|
Let errorSeti = the set of
instances in the training data misclassified by treei Let errorSetj = the set of
instances in the training data misclassified by treej |
|
|
Diversity = 1 - |errorSeti
intersection errorSetj| / |errorSeti union errorSetj| |
|
If a new tree does not satisfy the
minimum diversity requirement, the root is knocked off and another tree is
constructed. The maximal number of trials is specified by maxConsecutiveFails.
When the maxConsecutiveFails is reached, no new tree will be constructed and existing
trees will be used for classification.
To use other tree classifier than
J48, try
Classify -> Choose -> meta
-> MDMT