A Partial-Repeatability Approach to Data Mining

Unlike the data approached in traditional data mining activities, software data are featured with partial-repeatability or parepeatics, which is an invariant property that can neither be proved in mathematics nor validated to a high accuracy in physics, but still (partially) governs the behavior of the data. Parepeatics emerges as a result of the inaccurate universe. The universe comprises all possible C language programs is an example that cannot be accurately characterized since human writes defect-prone programs. In this paper we design a parepeatic mining framework for software data diming, where the mined knowledge is represented in terms of parepeatic models. A parepeatic model consists of central knowledge, a knowledge fluctuation zone and a correctness factor. Our approach can generate the required parepeatic model as a new form of knowledge representation from a given dataset and apply it to software data mining. Experimental results with real C language programs show that the proposed approach is effective.