Discernibility matrix simplification with new attribute dependency functions for incomplete information systems

摘要

Recently, many researches have been done on attribute dependency degree models. In this work, we bring forward three attribute dependency functions for incomplete information systems and investigate their basic properties in detail. Afterward, we apply the proposed models to twelve data sets from the UCI repository of machine learning databases. Finally, using the proposed functions, we perform the discernibility matrix simplification of incomplete information systems. The experimental results show that our proposed functions are more flexible to calculate the degree of each conditional attribute related to the decision attribute for incomplete information systems.