Identification of Reliable Information for Classification Problems

A novel information identification model is proposed to support accurate classification tasks with mixtures of categorical and real-valued attributes. This model combines the advantages of rough set theory and cluster validity method to promote the classification quality to the higher levels. Real-valued attribute values are pre-processed by fuzzy c-means clustering method and then analyzed by variable precision rough set theory. Our cluster validity index finalizes the information system with the feasible cluster number for each attribute. In the case that a considerable amount of ambiguous instances is included, the experimental results show that our model can explicitly improve traditional classifiers in the aspects of classification accuracy and discrimination power. This paper provides a better solution for the generation of reliable decision rules for classification problems with attribute mixtures. Key-Words: Reliable information, classification problems, fuzzy c-means, variable precision rough set, cluster validity index, discrimination power