| The rapid growth of all kinds of data in the field of education based on the development of educational information have prompted the research of educational data mining, which has become a hot spot. Educational data mining is a process which converts the raw data from a variety of educational system to knowledge which can be used by teachers, students, parents, education managers and education software system developers. The researches on educational data mining are late at home and abroad, and it is in the early stage of development at home. How to convert the stored data to knowledge and serve education decision-making process have become the concern of educational workers.This paper analyses the curriculum scores and IC card data generated by universities to establish a student status warning and decision support system. Knowledge base, data warehouse, data mining technology and multidimensional analytical processing are used to analyze and process the data, and realize the framework of the student status warning and decision support system. The system extracts information of students from different data sources periodically and stores in the data warehouse after pretreatment, the stored data have been deeply analyzed and mined by data mining technology and multidimensional analytical processing, then the knowledge to the knowledge base is saved. The data mining technology analyses the data from the following aspects: students' classification and the characteristics of every classification; the influence factors of students' comprehensive scores and scores prediction, and the process warning is also contained. Multidimensional analytical processing combined with warning level rule algorithm takes multidimensional analysis to curriculum information data and displays the results and development trend of warning, which is named routine warning. The system realizes study warning and decision support by interactive query and displaying of mining results.According to the characteristics of this paper's data set, the CFSFDP(Clustering by Fast Search and Find of the Density Peaks) clustering algorithm and NB(Naive Bayes) classifier must be improved, so the NM-CFSFDP(CFSFDP based on Neighbor Distance Curve and Merging Clusters) and NBC-IBA(Bayesian Classifier based on Improved Bat Algorithm) algorithm are proposed.In order to improve the accuracy and versatility of CFSFDP algorithm, the NM-CFSFDP algorithm is proposed. According to the uneven distribution of educational data sets and the clusters are obvious and may have the characteristics of multi density peaks, the following work is carried out. Firstly, the new algorithm gives the density threshold automatically, the density threshold is determined by the change of the nearest neighbor distance curve. Secondly, NM-CFSFDP used CFSFDP algorithm which gives density threshold automatically, to cluster the data set and then merges the classes that can be merged. Through the contrast experiment and application, the NM-CFSFDP algorithm is more accurate than the CFSFDP in clustering of ordinary data set and multi density peak data set.In order to make the Bayesian classifier applied to the correlation analysis and improve the accuracy of the classifier, the NBC-IBA algorithm is proposed. Firstly, a weight is assigned to each attribute, which can weaken the assumption of conditional independence and make correlation analysis; Then, in order to avoid falling into local optimal solution and accelerating convergence, the Taboo search mechanism and the random perturbation operator are introduced to the improved bat algorithm; Finally, improved bat algorithm are used to give different weight to every attribute and weaken the hypothesis of independence by searching the weight value for every attribute automatically. The experimental and applicative results show that comparing with the traditional naive Bayesian and newest weighted Bayesian classification algorithm, this proposed algorithm has better and more accurate classifications. |
PDF Full Text Request