基于机器学习的大学生学业预警技术研究

To train high-quality compound talents is basic need of national and social development. Therefore, timely and accurate prediction for acadamic crisis student problems appearing in their growth process is critical to higher education. University information system accumulated large amounts of data contains a wealth of current students' individual and group behavior for students, which would help to find academic crisis students in colleges and universities to provide new ideas. The project will be carried out based on machine learning techniques to analyze the data of College and University’s Students and to propose methods of feature extraction and feature selection. Then, we will propose models by considering following aspects: using relational learning techniques based on real and virtual communities to verify the diffusion of student's inappropriate behavior among groups; exploring imbalanced multi-label learning techniques to relieve the negative effects of the imbalance class distribution of sample on multi-label learning model brings; extracting useful information from a plurality of weak views for constructing co-training multi-view learning techniques; Proposing a technology based on "digital footprints" to find academic problem student in order to provide a new way to better understand the behavior of academic crisis student of College. To carry out this project is expected to play a role in promoting the research of machine learning based educational data analysis/mining technologies.

国家和社会的发展需要高素质复合型人才，及时准确地发现大学生在成长过程中出现的问题对高等教育至关重要。目前高校信息系统累积的大量数据中蕴含着丰富的学生个体和群体行为模式，为发现高校学业危机问题学生提供新的思路。本项目将对真实高校数据进行分析，构建面向学业预警问题的特征提取和特征选择方法，在此基础上基于机器学习技术从以下几个方面对学业危机问题大学生建模：提出基于真实及虚拟学生社群发现的关系学习技术，从而验证学业问题在学生间的传染性；提出面向学业问题学生的类别不平衡多标记学习技术，从而缓解样本分布不平衡对多标记学习建模带来的负面影响；从学生数据的多个弱视图中提取有效信息，提出基于多个弱视图协同训练的多视图学习技术；提出基于校园内“数字脚印”的学业问题学生发现技术，为更好的理解学业问题学生行为提供新的途径；本项目的开展有望为基于机器学习的教育数据分析/挖掘技术研究起到一定的推动作用。

国家和社会的发展需要高素质复合型人才，及时准确地发现大学生在成长过程中出现的问题对高等教育至关重要。目前高校信息系统累积的大量数据中蕴含着丰富的学生个体和群体行为模式，为发现高校学业危机问题学生提供新的思路。本项目对真实高校数据进行分析，截至2021年12月，项目组提出面向学业危机预警的特征选择方法；提出学生社群发现的机器学习技术；提出面向学业问题学生的类别不平衡多标记学习技术；提出基于弱视图协同训练的多视图学习技术；提出基于轨迹数据挖掘的学业问题学生发现技术；在重要国际期刊、国际会议及国内一级学报上发表11篇高质量论文；培养研究生10名，其中毕业5名（博士1名，硕士4名）。