Research And Application Of Attribute Reduction And Weighting Method

In recent years,with the rapid increase of massive data,data mining has become a widely applied technology.In the massive data processing,the research on attribute reduction method is a hot topic for many researchers.How to recognize the important attributes from the high-dimensional data effectively is one of the most important task in machine learning.In the process of modeling and decision making,the research on attribute weighting method is the inevitable issue.How to reasonably allocate the weigh values of the different attributes is the key to improve the accuracy and the level of decision making.This dissertation focuses on some shortcomings in the existed attribute reduction algorithm and attribute weighting algorithm,respectively,thereby proposing the improved editions.And these improved algorithms are applied to the prediction of liver cancer microvascular invasion.Firstly,a neighborhood rough set based attribute reduction algorithm with chi-square test is proposed.It’s used to solve the problem that the interaction between related attributes is not considered in the neighborhood rough set based attribute reduction algorithm.Firstly,calculate the correlation between attributes with chi-square test method,then consider the effect between related attributes in attribute reduction.That is,calculate the sum of the importance of a single attribute and its associated attributes,which makes the screening results more accurate and effective.Experiments show that the algorithm has good effect.Secondly,an attribute weighting algorithm combining subject and object is proposed.It’s aimed at the problem that attribute weighting algorithm based on analytic hierarchy process is too subjective in constructing judgment matrix.Firstly,calculate the coefficients between attributes with Pearson correlation coefficient.Then construct the pair judgment matrix according to the coefficient size,and calculate the attribute weights,and obtain the final attribute weight value by combining the weight value with the objective analysis of entropy weight.Comparing to the completely subjective matrix construction,the improved method avoids uncertainty to a certain extent.It can not only reflect the subjective empirical knowledge judgement,but also excavate the potential value of objective data.The attribute reduction algorithm and attribute weighting algorithm proposed in this thesis are applied to the prediction of liver cancer microvascular invasion.Firstly,the attribute reduction algorithm is used to reduce the dimension of the data of liver cancer microvascular invasion,then calculate the weight of each attribute by using the attribute weighting algorithm proposed in this thesis.Finally,it is combined with the gradient boosting tree classification model to construct the prediction model of liver cancer microvascular invasion.from the accuracy,sensitivity,specificity and the results of the receive operation curve of the prediction model,the proposed algorithm and the classification model have better combination effect,and all aspects of the index have achieved better effect.