System Uncertainty Based Data-Driven Knowledge Acquisition

In the three-layered framework for knowledge discovery, it is necessary for technique layer to develop some data-driven algorithms, whose knowledge acquiring process is characterized by and hence advantageous for the unnecessity of prior domain knowledge or external information. System uncertainty is able to conduct data-driven knowledge acquiring process. It is crucial for such a knowledge acquiring framework to measure system uncertainty reasonably and precisely. Herein, in order to find a suitable measuring method, various uncertainty measures based on rough set theory are comprehensively studied: their algebraic characteristics and quantitative relations are disclosed; their performances are compared through a series of experimental tests; consequently, the optimal measure is determined. Then, a new data-driven knowledge acquiring algorithm is developed based on the optimal uncertainty measure and the Skowron’s algorithm for mining propositional default decision rules. Results of simulation experiments illustrate that the proposed algorithm obviously outperforms some other congeneric algorithms.