الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 151 |  |  |
| | | from | 151 | | |
Abstract
Data mining is the science of discovering unknown and interesting patterns from databases. One approach to discover knowledge from data is to learn classification knowledge. Classification knowledge can be identified as descriptive generalization of portions of the space of all possible data points. Almost all classification algorithms utilize attribute selection criteria for reaching the best generalizations. Attribute selection criteria are biased by their mathematical model. Classification algorithms employee different alorithms to determine how much generalization is enough for each portion of the space.
Decision tree algorithms are symbolic classification models that partition the space of all possible outcomes based on the classification relationship between each attribute of the given data and the decision attribute is measured by an attribute selection criterion.Attribute selection criterion may work well with some data, and may not work well with other data. This thesis investigates utilizing multiple attribute selection criteria for building a single decision tree.
This work develops a new system known as multi-criteria decision tree MCDT learning system. it allows the decision tree to be learned using a combination of three attribute selection critria: gain ratio, chi-square x2 and apriori, then the learned tree is pruned using expected error pruning algorithm. The user utilizes a parameter to adjust the pruning process to control the level where the pruning process takes place. The predictive accuracies of the decision tree learned using each of the tree attribute selection criteria are calculated and compared with the new approach for fourteen data sets.
The obtained decision tree is visualized to the user, the visualized decision tree can be relearned with more than one attribute selection criteria. The user can modify the decision tree at any node by selecting a different attribute selection criteria, and reconstructing the subtree branches from the selected node. This process can be repeated anu number of times using multiple criteria for learning one decision tree.