Research On The Prediction Of Students' Achievement In Educational Data Mining By Cooperative Filtering Algorithm

Currently, the education of students in the school accumulated a lot more obvious variety of data, such as student enrollment, dropout rates and student achievement scores of data subjects, in particular the right to rate their classes to answer the question of knowledge mastery of information. Clearly, these various types of data in the field of education is constantly changing, as will the development of information and accumulation increased, how to extract these complex data burdensome useful information, with good research value.Combining collaborative filtering algorithm similarity in the field of e-commerce and other data analysis, collaborative filtering algorithm will be applied to the data in the field of education, focusing on student achievement prediction research on KDD Cup 2010 Competition selected from ITS Intelligent Tutor System 8.9 million pieces of data as the experimental data sets were student achievement data mining prediction education practice. Experimental data set features a large amount in the range is large, mostly text-type data, partial data sparse and so on. To solve these problems, this paper carried out the following work:(1) an incremental sampling method, to determine the optimal size of the training set, substantially reduce the amount of training set records; combined data set time characteristics, the training set to extract the latest data of N; remove the implicit answer large result sets null proportions feature, part of a complex structure of separate property.(2) a single K-nearest neighbor classification algorithm and singular value decomposition SVD model apply to educational data set to validate the test set Correct First Attempt (CFA) property prediction, and as evaluation of the content, while comparison of the two algorithms prediction.(3) This article is also based on two base algorithms complementary characteristics, the SVD dimensionality reduction and K-nearest neighbor algorithm combined forecast student achievement. Experiments can be analyzed, the algorithm enables data sparsity eased to some extent, but only to retain the basic characteristics of the data, partial data loss caused by dimensionality reduction evaluation results will cause little impact.