Title

Artificial Intelligence – I: A two-step approach for improving efficiency of feedforward Multilayer Perceptrons network

Abstract/Description

An artificial neural network has got greater importance in the field of data mining. Although it may have complex structure, long training time, and uneasily understandable representation of results, neural network has high accuracy and is preferable in data mining. This research paper is aimed to improve efficiency and to provide accurate results on the basis of same behaviour data. To achieve these objectives, an algorithm is proposed that uses two data mining techniques, that is, attribute selection method and cluster analysis. The algorithm works by applying attribute selection method to eliminate irrelevant attributes, so that input dimensionality is reduced to only those attributes which contribute in the training process. Then after, the whole dataset is partitioned into n clusters which are finally fed into multilayer perceptrons network based on backpropagation algorithm to carry out blockwise and parallel training.

Session Theme

Artificial Intelligence – I

Session Type

Other

Session Chair

Dr. Sajjad Haider

Start Date

15-8-2009 3:25 PM

End Date

15-8-2009 3:45 PM

Share

COinS
 
Aug 15th, 3:25 PM Aug 15th, 3:45 PM

Artificial Intelligence – I: A two-step approach for improving efficiency of feedforward Multilayer Perceptrons network

An artificial neural network has got greater importance in the field of data mining. Although it may have complex structure, long training time, and uneasily understandable representation of results, neural network has high accuracy and is preferable in data mining. This research paper is aimed to improve efficiency and to provide accurate results on the basis of same behaviour data. To achieve these objectives, an algorithm is proposed that uses two data mining techniques, that is, attribute selection method and cluster analysis. The algorithm works by applying attribute selection method to eliminate irrelevant attributes, so that input dimensionality is reduced to only those attributes which contribute in the training process. Then after, the whole dataset is partitioned into n clusters which are finally fed into multilayer perceptrons network based on backpropagation algorithm to carry out blockwise and parallel training.