Please use this identifier to cite or link to this item: http://hdl.handle.net/10603/17452
Title: Novel Algorithms for Knowledge Discovery from Neural Networks in Classification Problems
Researcher: M. Gethsiyal Augasta
Guide(s): Dr. T.KATHIRVALAVAKUMAR
Keywords: computer, novel, algorithm, discovery, neural network
Upload Date: 12-Mar-2014
University: Mother Teresa Womens University
Completed Date: 17/06/2013
Abstract: Large datasets encompass hidden trends which convey valuable knowledge about the dataset. Data mining research deals with extraction of useful and valuable information from such large datasets. The process of data mining can be viewed as exploration and analysis of large quantities of data, by automatic or semiautomatic means, in order to discover meaningful patterns and rules. One of the newlinemost important function of data mining is classification. It recognizes patterns that newlinedescribe the group to which an item belongs. It does this by examining existing newlineitems that already have been classified and inferring a set of rules. Artificial neural newlinenetworks have been widely used to develop highly accurate classifiers for the realworld newlineproblem domains using different learning algorithms. newlineEventhough there exists a lot of learning algorithms for neural networks to resolve newlinedifferent types of problems, still the artificial intelligence incorporated in the newlineneural network is only to the level of tapeworm. Researches are going, in different newlinedirections by finding new preprocessing methods, topology and rule extraction algorithms to maximize the classification accuracy and to extract the knowledge in newlinethe form of rules to classify the real life complex problems. In this research, focus newlinehas been given to overcome the problems faced in classification using neural networks, newlineand new novel algorithms have been proposed for the success of feedforward neural networks on classification problems. newlineThe thesis first proposes discretization algorithms for preprocessing the data for neural networks classifier. The proposed algorithms discretize the data based on newlinethe mean value / the range coefficient of dispersion and skewness. They automate newlinethe discretization process by computing the number of intervals and stopping criterion. newlineThe Backpropagation(BP) with momentum training algorithm and conjugate gradient training algorithm are used to compute the accuracy of classification on feedforward neural network from thedata discretized by these algorithms
Pagination: 178
URI: http://hdl.handle.net/10603/17452
Appears in Departments:Department of Computer Science

Files in This Item:
File Description SizeFormat 
01_title.pdfAttached File53.3 kBAdobe PDFView/Open
02_certificate.pdf7.96 kBAdobe PDFView/Open
03_abstract.pdf93.46 kBAdobe PDFView/Open
04_declaration.pdf8.12 kBAdobe PDFView/Open
05_acknowledgement.pdf88.51 kBAdobe PDFView/Open
06_contents.pdf92.89 kBAdobe PDFView/Open
07_list_of_tables.pdf52 kBAdobe PDFView/Open
08_list_of_figures.pdf94.21 kBAdobe PDFView/Open
09_chapter 1.pdf393.36 kBAdobe PDFView/Open
10_chapter 2.pdf458.39 kBAdobe PDFView/Open
11_chapter 3.pdf415.57 kBAdobe PDFView/Open
12_chapter 4.pdf411.09 kBAdobe PDFView/Open
13_chapter 5.pdf462.78 kBAdobe PDFView/Open
14_chapter 6.pdf736.33 kBAdobe PDFView/Open
15_conclusion.pdf111.16 kBAdobe PDFView/Open
16_bibliography.pdf169.56 kBAdobe PDFView/Open


Items in Shodhganga are protected by copyright, with all rights reserved, unless otherwise indicated.