An exploration of neural network and decision tree approaches for knowledge discovery in databases

Pao Yue-kong Library Electronic Theses Database

An exploration of neural network and decision tree approaches for knowledge discovery in databases

 

Author: Lui, Chun-man
Title: An exploration of neural network and decision tree approaches for knowledge discovery in databases
Degree: M.Sc.
Year: 1997
Subject: Database management
Neural networks (Computer science)
Decision trees
Knowledge acquisition (Expert systems)
Hong Kong Polytechnic University -- Dissertations
Department: Multi-disciplinary Studies
Pages: v, 167 leaves : ill. ; 30 cm
Language: English
InnoPac Record: http://library.polyu.edu.hk/record=b1250467
URI: http://theses.lib.polyu.edu.hk/handle/200/2005
Abstract: Classification and regression are two of the most popular data mining tasks. Many decision-making processes fall into the general category of classification. Two important categories of empirical learning techniques for classification are neural networks and machine learning algorithms for induction of decision trees or production rules. The two categories differ radically in their underlying models and the representation of their solution. A decision tree approach offers a clearly explained format for a decision, while the acquired solution from a neural network is difficult to be explained. Experimental comparisons of these two categories of learning algorithms have been reported in the literature. In this study, the neural network and decision tree approaches were further explored, and applied to three real-world datasets. In particular, the C4.5 and OC1 systems were used in the decision tree method, while the backpropagation learning was used in the neural network approach. The classification accuracy of these learning systems were then compared. An information theoretic class-dependent discretization method (CADD) was implemented to transform the continuous attributes into ordered discrete ones, with a view to improve C4.5. Training and testing or cross-validation techniques were used to estimate the error rates of the classification methods. Detailed analysis of performance of the neural networks was also performed. The results indicated that the C4.5 decision tree classifier using data discretized by the discretization algorithm improved its classification accuracy. In terms of the proportion of all cases correctly classified, the performance of the backpropagation neural network was the best, while the OC1 learning system outperformed the C4.5 system. As another issue, since neural networks have been criticized for "not being able to explain their conclusions", a method for extracting rules from neural network and facts (the RN method) was implemented to generate a set of If-Then rules in Disjunctive Normal Form. Finally, as an exploration of the power of backpropagation in predicting real-valued attributes, a neural network was trained to perform a forecasting task in connection with meteorology.

Files in this item

Files Size Format
b12504671.pdf 3.958Mb PDF
Copyright Undertaking
As a bona fide Library user, I declare that:
  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

     

Quick Search

Browse

More Information