Knowledge discovery is a new research field rising in the early 1990 s. It is currently a hot issue in the field of artificial intelligence and data mining. Knowledge discovery is to extract the abstract, comprehensible essence and contact of objective world which is previously unknown, using a variety of methods like statistics, rough set, fuzzy mathematics, machine learning and expert systems, etc. from a large number of incomplete, noisy, fuzzy and random practical application data.Rough set theory is an efficient mathematical tool dealing with imprecise,incomplete and inconsistent data. It has already made great strides in its theory and application. This dissertation mainly studies several types of knowledge discovery model based on rough set theory and its reduction algorithm, and applies the algorithm into the automatic analysis of ECG signal diagnosis.The main research contents are as follows:1. The basic theory of rough set and reduction algorithm is discussed. An improved definition of approximation accuracy is proposed. The new definition overcomes the insufficiency of Pawlak approximation accuracy, which does not reflect the influence of knowledge granularity on the approximation accuracy. An improved reduction algorithm based on knowledge dependence is proposed. The improved algorithm has some advantages of less time complexity and lower algorithm complexity. A reduction algorithm of inconsistent information system towards the small sample problem is proposed. This algorithm avoids the loss of useful information due to the insufficient sample data in some practical application, like data in medical clinical diagnosis and UCI machine learning data. The experiments show its effectiveness.2. Granular computing model based on rough set is researched. Some important properties of granular computing and the reduction algorithms are discussed. An improved reduction algorithm based on attribute significance of granular computing is proposed. The traditional reduction algorithm based on granular computing is gradually calculated with reduction, but some information systems may have no reduction core. In this case, this thesis has made an improvement to reduction algorithm, and it causes the algorithm can be used in system with both reduction coreand no reduction core, thus, the feasibility of it is approved through experiments.3. A new cognitive model based on attribute granular is presented. The formation mechanism of cognitive process is researched. The properties of the cognitive information granular and its application are discussed. Then a reduction algorithm combined with knowledge dependence and attribute significance is proposed.4. This paper shows the electrocardiograph(ECG) automatic analysis as the application background of attribute reduction. In chapter five, a certain work has been done to the definition of characteristic parameters and feature selection by rough set.The algorithm is validated by the experimental result.
