高校学生日常行为异常检测的一种改进模糊聚类方法

发布时间：2018-03-09 10:48

本文选题：学生行为　切入点：异常检测　出处：《大连海事大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着数字化和信息化技术在各大高校中的广泛普及,智能管理系统得到了广泛的应用和高度重视,是高校信息化程度的重要的标志。伴随着一卡通的应用,高校里的各类管理信息系统积累了大量的数据,如学生成绩、消费信息、上网信息、图书馆借阅信息、进出宿舍楼情况等等。但我们对数据处理和应用,仍然只是停留在一个简单的查询和统计报告阶段,并没有采取有效的措施,充分利用这些数据,为学生管理提供有效的决策支持。而在这些大量数据中,对异常数据的研究比正常数据往往更有意义。因此本文的重点就是利用这些数据对学生的行为进行分析,找出行为异常的学生,为后续学校的管理工作提供决策支持。本文首先研究现阶段对异常研究领域的一些重要方法,如:基于统计分布的异常检测方法、基于距离的异常检测方法、基于偏差的异常检测方法、基于深度的异常检测方法、基于密度的异常检测方法、基于聚类的异常检测方法,并分析了这些常用的异常检测算法的优劣度。其次,本文从学生行为的概念入手,在前人研究的基础上,介绍了本文将要着重研究的学生行为:消费行为、课程成绩、正常回寝次数、上网时间等等。这些行为将作为后期对学生的行为特征进行异常检测的四个维度,对学生的行为异常进行一个鉴定。在本文中异常并无褒义贬义之分,这里的异常概念只是区分个别学生与大部分学生的一个不同。利用特征数据对学生行为进行表示。通过对基于距离的异常检测算法的研究,提出了基于距离的改进的相似度和的算法。随后通过对聚类思想的分析,发现聚类和异常检测在某种程度上存在相似之处,通过分析模糊聚类算法的缺点,提出利用改进的模糊C-均值聚类算法对异常进行检测。最后通过实验对两种算法的结果进行对比分析。
[Abstract]:With the wide spread of digitization and information technology in colleges and universities, intelligent management system has been widely used and highly valued, which is an important symbol of the degree of information in colleges and universities. All kinds of management information systems in colleges and universities have accumulated a lot of data, such as student achievement, consumption information, Internet information, library borrowing information, entering and leaving the dormitory building, etc. But we process and apply the data. Still remain at a simple query and statistical reporting stage, and have not taken effective measures to make full use of these data to provide effective decision support for student management. The study of abnormal data is often more meaningful than that of normal data. Therefore, the emphasis of this paper is to use these data to analyze students' behavior and to find students with abnormal behavior. This paper first studies some important methods in the field of anomaly research, such as: anomaly detection method based on statistical distribution, anomaly detection method based on distance. Anomaly detection method based on deviation, anomaly detection method based on depth, anomaly detection method based on density, anomaly detection method based on clustering, and the advantages and disadvantages of these commonly used anomaly detection algorithms are analyzed. Starting with the concept of student behavior and on the basis of previous studies, this paper introduces the student behavior that this paper will focus on: consumer behavior, course achievement, normal bedtime, These behaviors will be used as four dimensions of abnormal detection of students' behavioral characteristics in the later stage, and will be used to identify the abnormal behaviors of students. In this paper, there is no positive or derogatory meaning of abnormal behavior. The concept of anomaly here is only a difference between individual students and most students. Using feature data to express students' behavior, we study the distance-based anomaly detection algorithm. An improved similarity sum algorithm based on distance is proposed. Then, by analyzing the clustering idea, it is found that there are similarities between clustering and anomaly detection to some extent, and the shortcomings of fuzzy clustering algorithm are analyzed. An improved fuzzy C-means clustering algorithm is proposed to detect anomalies. Finally, the results of the two algorithms are compared and analyzed through experiments.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13

【相似文献】