Title: DECISION TREES ALGORITHMS AND CLASSIFICATION WITH INCREMENTAL NEURAL NETWORK

Issue Number: Vol. 5, No. 3
Year of Publication: Jul - 2015
Page Numbers: 203-209
Authors: John Tsiligaridis
Journal Name: International Journal of Digital Information and Wireless Communications (IJDIWC)
- Hong Kong
DOI:  http://dx.doi.org/10.17781/P001710

Abstract:


The purpose of this paper is to present a set of algorithms for rule extraction using direct and indirect methods and the use of Neural Network for classification. A direct method with simplification (DMS) to extract rules from data using a hash table is also developed. Apart from the rules, DMS can also produce the corresponding decision tree for each subset with predefined class value. The merge of the paths of all call values can produce the decision tree of the data. Therefore, DMS works as both direct and indirect method. For the indirect method a Decision Tree algorithm (DTA) is created from data using probabilities, aiming at creating on-demand an accurate decision tree (DT) from either data or a stable set of rules. The advantages of both of them to avoid redundant branches are presented. DMS is the ability to produce rules with a maximum number of conditions and prevention of subtree overlapping. Moreover, an incremental Backpropagation Neural Network (IBNN) is created using the instances of data grouping according to their predefined class. Comparisons show that IBNN outperforms DMS. Simulation results are provided.