基于随机森林的用户对在线课程的放弃预测  

Based on Random Forest Users Online Course Waiver Forecast

作　　者：王晓杰[1] 孙仁诚[2] 邵峰晶[1,2] 

机构地区：[1]青岛大学计算机科学技术学院,山东青岛266071 [2]青岛大学自动化与电气工程学院,山东青岛266071

出　　处：《青岛大学学报（工程技术版）》2016年第4期17-21,共5页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：国家自然科学基金资助项目(91130035);国家公益性行业科研专项基金资助项目(200905030-2);山东省自然科学基金资助重点项目(ZR2012FZ003);山东省自然科学基金资助项目(ZR2012FQ017)

摘　　要：针对大规模在线开放课程选择的自主性而导致的课程辍课率高的问题,本文根据中国MOOC平台学堂在线的数据,采用随机森林建模方法,通过分析MOOC平台上学习者的学习日志数据,找出学习者学习课程的模式规律,建立课程学习行为预测模型,并通过特征选择,选取6个特征进行模型预测。预测结果表明,由于特征值的均值差异较大,所以这6个特征对预测结果有很大的影响;将删选后的方法与方法1进行秩和检验,发现P值非常小,接近于0,表明使用6个重要性强的特征对用户进行预测的方法是5个方法中最好的,这样既使用较少的特征,又有较好的性能。因此,在进行课程预测时,无需构建大量的特征进行建模,还能提高模型的训练速度,获得精确度高的预测结果。In order to design reasonable static and dynamic output feedback controllers for singular Delta operator systems to ensure that the corresponding closed-loop systems are admissible and have a prescribed H_∞ performance indexγ,this paper deals with the problem of output feedback H_∞control for singular Delta operator systems.By using the method of matrix inequalities,necessary and sufficient conditions for the existence of suitable static and dynamic output feedback H_∞ controllers are presented,respectively.Moreover,the design methods of the corresponding controllers are obtained based on the solutions to matrix inequalities and a numerical example is given to demonstrate the theoretical results in this paper.The results showed that the designed static and dynamic output feedback controllers according to the method in this paper can indeed ensure that the closed-loop systems are admissible and have H_∞ performance indexγ.Thus,the controller design methods in this paper are indeed effective and feasible.The study has practical significance.

关 键 词：MOOC 特征删选 数据挖掘 随机森林 

分 类 号：O212.1[理学—概率论与数理统计]