Decision Tree Rule Reduction Using Linear Classifiers in Multilayer Perceptron

It hasbeenshownthat a neuralnetworkis better thana direct applicationof inductiontrees in modelingcomplex relations of inputattributes in sampledata. We proposethat conciserules be extractedto supportdata withinputvariablerelations overcontinuous-valuedattributes. Thoserelationsas a set of linear classifiers can be obtained fromneural networkmodelingbased onback-propagation.Alinear classifier is derivedfrom a linear combinationof input attributes and neuron weightsin the first hiddenlayerof neuralnetworks.It is shownin this paperthat whenweuse a decisiontree overlinear classifiers extractedfroma multilayerperceptron, the numberof rules canbe reduced.Wehave tested this methodoverseveraldatasets to compareit withdecisiontree results.

• Type: Conference
• Year: 2001
• Where: FLAIRS
• Authors: DaeEun Kim, Sea Woo Kim
• Keywords: Artificial Intelligence, Decisiontree Overlinear Classifiers, FLAIRS 2001, Linear Classifiers, Linear Combinationof Input