Research On Application Of Data Mining In College Scale Analysis And Decision

With the rapid development of computer technology, the demand fordepth analysis of the data is gradually increased, and data miningtechnology has been widely applied. With the reality of our country, theministry of education set the index of teaching administration housing areaof each student in university evaluation index system. Whether the index isreasonable or not, it should adopt effective data mining methods to analysethe large amount of historical data, and provide relevant basis for thedeciders of college and education department to make full use of resources.First, the concept, process and method etc of data mining technologyare deeply analyzed, and the typical model, various learning rules ofartificial neural network theory are carefully studied and summarized. Thethesis discusses the mathematical model, learning algorithm and thestructure design of BP neural network. It establishes the theoreticalfoundation for the construction of the model.Secondly, according to the index of teaching administration housing area of each student in college scale, nearly10years of college related datafrom Liaoning province is collected. The LOGISTIC model of the collegescale is built by the data mining tool which called Enterprise miner in SASsoftware. The index of teaching administration housing area of each studentis checked by the LOGISTIC model, finding that the index is unreasonable.Another model is built by using BP neural network in the SAS/EM. Thereis different impact for the result of network model with different networkhidden nodes and different learning rate. The model with the least meansquare value of validation data is chose as the optimal BP neural networkmodel. The colleges are divided into three groups, and the index ofteaching administration housing area of each student for the three groups isforecasted by the optimal BP neural network model.Finally, the predicted value of the index is calculated by training theBP neural network model. In order to improve the population of the modelfor other data sets expect the training sample, the network model is revisedby using two kinds of methods. The model revised by Early Stoppingmethod has the least total loss, and it can be used for the promotion of themodel.