A New Heuristic Algorithm of Rules Generation Based on Rough Sets

Generating decision rules is one of the most important data mining areas which ldquorough set data analysis(RSDA)rdquo can address. Generally, for the same expression, the shorter the rules are, the more effectively the system performances. Considering of this, this paper provides a new heuristic algorithm named ldquoshort first extraction (SFE)rdquo based on the classical rough set theory, for rules generation. A standard named ldquoall attribute in rulespsila length(AARL)rdquo to compare the rulespsila ability is also provided. Our experiments is based on the datasets provided by UCI machine learning repository, such as iris datasets, new-thyroid dataset and yellow-small(balloons) dataset. The experimentspsila results indicate that ldquoSFErdquo always has better performance than JohnsonReducer, genetic reducer and Holtepsilas 1R reducer: it always generates less rules and has lower ldquoAARLrdquo than its competitors. Our ldquoSFErdquo algorithm also has another property which may be useful: the rules generated by ldquoSFErdquo is a covering but not a partition of the information system, and it may lead us to a new direction of rules generating research.