The BP Neural Network Based On Rough Sets And Personal Credit Evaluation Model

With the rise of big data, data mining methods that people require are increasingly high. Now the database only for information records and management resources, not for the perspective of the collection of data mining that cause data uncertainty, incompleteness, redundancy, etc, traditional data mining methods can’t meet the requirements. Rough set is a method and data analysis tool of dealing with inconsistent, uncertain, vague powerful mathematical, without any prior knowledge. Attribute reduction is one of the core of rough set theory, you can keep the same classification ability to reduce the amount of data in the database, thereby reducing the space and time complexity of data mining. This paper studies the classical attribute reduction algorithm and contrary to their shortage presents an improved reduction algorithm based on a new attribute importance; This paper also studies the combination of rough sets and BP neural network, namely rough set attribute reduction as a front-end processor. From BP neural network theory, neural network structure is simplified, shorten training time, and finally through the empirical analysis of individual credit assessment to verify the effectiveness of the algorithm. The main contents of this paper are as follows:Firstly,analysis of the research rough sets, data mining and combine rough sets with data mining; in-depth study of the basic theory of rough sets, such as rough set theory approximation set down, domain knowledge, reduction concepts and study the basic data mining technologies as data mining processes, methods and tasks.Secondly,in-depth study of the classical rough set attribute reduction algorithm and analysis of their respective strengths and weaknesses, thereby improved algorithm is proposed. For complex logical formulas of calculation method to computing core based on the distinction function that proposed computing core based on improved distinguish matrix; for attribute importance(dependence, information entropy) incompleteness that defined a new attribute importance, considering the positive impact on the region and the border region reduction process; For the problem that can’t be distinguished attribute importance attribute that used VPRS strategy for equal importance to distinguish between the properties again, this paper propose a property based on a new attribute importance about reduction algorithm, and select five groups of discrete data sets from UCI database for simulation tests and demonstrate the effectiveness and feasibility of the algorithm.Finally,stduy of BP neural network structure and algorithm process; In this paper, combine rough sets with neural network due to their respective advantages and disadvantages. Useing reduction algorithm based on new attribute importance as a new front-end processor for network and through individual credit assessment model empirical analysis of the algorithm, compared no reduction personal credit classification that accuracy decline slightly, but after reduction, input vector dimension is greatly reduced, BP neural network structure is simplified and training time is reduced, the model is extended to large decision tables and data set has practical significance.