Using The Set of Attributes of Frequent Itemsets for Better Rough Set based Rules

摘要

Inductive learning algorithms want to find a function that reflects a given sequence of input and output pairs where the input and output pairs consist of value vectors. Rough set systems can extract minimal set of rules that act as the function of inductive learning, and are well known for their strong mathematical background. The found set of rules is solely based on data so that no prejudiced views can be inserted in the found rule set. But even though the good property, rough set based rule systems have the tendency of being unstable in the sense that their performance is very dependent on given training data sets due to their sole reliance on given data. In order to avoid such property of the rough set based rule systems this paper suggests using the attributes of frequent items only in the input vector to find the rules. The attributes can be found by applying association rule algorithms. Experiments with several real world data sets show that better rough set based rules could be found in accuracy by using the attributes only, especially when the attributes of input have key-like characteristics.