WILS：面向学业预警的非均衡增量式学习方法

doi:10.7523/j.ucas.2021.0055

中国科学院大学学报 ›› 2023, Vol. 40 ›› Issue (3): 422-432.DOI: 10.7523/j.ucas.2021.0055

• 简报 • 上一篇

WILS：面向学业预警的非均衡增量式学习方法

盛晓光¹, 王颖¹, 张迎伟^1,2, 项若曦³, 付红萍^4,5

1. 中国科学院大学人工智能学院, 北京 100049;
2. 中国科学院计算技术研究所, 北京 100190;
3. 北京语言大学信息科学学院, 北京 100083;
4. 北京林业大学信息学院, 北京 100083;
5. 国家林业和草原局林业智能信息处理工程技术研究中心, 北京 100083

收稿日期:2021-01-22 修回日期:2021-04-27 发布日期:2021-04-27
通讯作者: 项若曦,E-mail:rxxiang@blcu.edu.cn
基金资助:
国家自然科学基金（61702038）和中国科学院大学管理支撑创新能力提升专项（E0E58914）资助

A weighted incremental learning scheme for effective academic warning

SHENG Xiaoguang¹, WANG Ying¹, ZHANG Yingwei^1,2, XIANG Ruoxi³, FU Hongping^4,5

1. School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 100049, China;
2. Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China;
3. School of Information Science, Beijing Language and Culture University, Beijing 100083, China;
4. School of Information Science and Technology, Beijing Forestry University, Beijing 100083, China;
5. Engineering Research Center for Forestry-oriented Intelligent Information Processing, National Forestry and Grassland Administration, Beijing 100083, China

Received:2021-01-22 Revised:2021-04-27 Published:2021-04-27

摘要/Abstract

摘要： 学业预警是构建完善教育管理体系的基础，有助于及早发现、干预学生学习和生活中的异常状况。然而实际研究仍面临诸多挑战，如：1)学生学业表现的相关影响因素往往不断变化，导致数据分布的变化；2)学业预警数据集一般存在类别不均衡的问题。针对上述挑战，提出一种面向学业预警的非均衡增量式学习方法(WILS)。WILS由增量学习机制和加权损失函数两部分构成，增量学习机制能够适应数据分布和样本特征的动态变化，加权损失函数通过为少数类赋予更高的权重提升对该类别的关注度。为评估WILS的效果，在包含2 275名本科生的真实数据集和包含1 000名学生的公开数据集上进行了实验验证。结果表明，相较于已有方法，WILS能够较好地适应数据和特征的连续动态变化，取得优异的识别效果。

关键词: 学业预警, 增量学习, 非均衡, 深度学习, 加权损失函数

Abstract: Academic warning plays an essential role in delivering personalized intervention for the student early and constructing a sound educational management system. However, practical explorations are usually limited by two challenges:1) Many influenced factors (e.g., curriculum setting) vary from the time, causing the change of data distribution; 2) The abnormal samples are generally rare compared with normal samples, and the available dataset tends to be imbalanced. To handle the challenges above, this paper proposes a Weighted Incremental Learning Scheme, namely WILS, to realize the effective academic warning of undergraduates. WILS consists of two essential components:1) An incremental learning mechanism that supports the changes of data distribution and features setting; 2) A weighted loss function that will assign a higher weight to the abnormal category. To evaluate the effectiveness of WILS, we conduct experiments on real dataset with 2 275 undergraduates and public-available dataset with 1 000 students. Experimental results demonstrate that WILS is significantly more accurate and efficient compared with other methods.

Key words: academic warning, incremental learning, imbalance, deep learning, weighted loss function

中图分类号:

TP391

盛晓光, 王颖, 张迎伟, 项若曦, 付红萍. WILS：面向学业预警的非均衡增量式学习方法[J]. 中国科学院大学学报, 2023, 40(3): 422-432.

SHENG Xiaoguang, WANG Ying, ZHANG Yingwei, XIANG Ruoxi, FU Hongping. A weighted incremental learning scheme for effective academic warning[J]. Journal of University of Chinese Academy of Sciences, 2023, 40(3): 422-432.

参考文献

[1] Yao H X, Lian D F, Cao Y, et al. Predicting academic performance for college students:a campus behavior perspective[J]. ACM Transactions on Intelligent Systems and Technology, 2019, 10(3):1-21.
[2] Deary I J, Strand S, Smith P, et al. Intelligence and educational achievement[J]. Intelligence, 2007, 35(1):13-21.
[3] Taras H, Potts-Datema W. Obesity and student performance at school[J]. Journal of School Health, 2005, 75(8):291-295.
[4] White K R. The relation between socioeconomic status and academic achievement[J]. Psychological Bulletin, 1982, 91(3):461-481.
[5] Marston D B. A curriculum-based measurement approach to assessing academic performance:what it is and why do it[M]//Shinn M R. Curriculum-based measurement:assessing special children. New York:Guliford Press, 1989:18-78.
[6] Muthukrishnan S M, Govindasamy M K, Mustapha M N. Systematic mapping review on student's performance analysis using big data predictive model[J]. Journal of Fundamental and Applied Sciences, 2018, 9(4S):730-758.
[7] Liu L L, Yu S N. The academic warning model based on TLBO-BP algorithm[C]//201814th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery (ICNC-FSKD). July 28-30, 2018, Huangshan,China. IEEE, 2018:65-70.
[8] Fei M, Yeung D Y. Temporal models for predicting student dropout in massive open online courses[C]//2015 IEEE International Conference on Data Mining Workshop (ICDMW). November 14-17,2015,Atlantic City,NJ,USA. IEEE, 2015:256-263.
[9] Sun C, Xia F, Wang Y, et al. A deep learning model for automatic evaluation of academic engagement[C]//Proceedings of the Fifth Annual ACM Conference on Learning at Scale. London United Kingdom.New York,NY,USA:ACM, 2018:1-4.
[10] Yin S J, Shang Q D, Wang H Z, et al. The analysis and early warning of student loss in MOOC course[C]//Proceedings of the ACM Turing Celebration Conference-China. Chengdu,China.New York,NY,USA:ACM, 2019:1-6.
[11] Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]//2017 IEEE International Conference on Computer Vision (ICCV). October 22-29,2017,Venice Italy. IEEE, 2017:2999-3007.
[12] Gonzalez E C, Hernandez E C, Coltrane A K, et al. The correlation between physical activity and grade point average for health science graduate students[J]. OTJR, 2014, 34(3):160-167.
[13] Muñoz-Bullón F, Sanchez-Bueno M J, Vos-Saz A. The influence of sports participation on academic performance among students in higher education[J]. Sport Management Review, 2017, 20(4):365-378.
[14] Robbins S B, Lauver K, Le H, et al. Do psychosocial and study skill factors predict college outcomes? A meta-analysis[J]. Psychological Bulletin, 2004, 130(2):261-288.
[15] Zeek M L, Savoie M J, Song M, et al. Sleep duration and academic performance among student pharmacists[J]. American Journal of Pharmaceutical Education, 2015, 79(5):63.
[16] Laidra K, Pullmann H, Allik J. Personality and intelligence as predictors of academic achievement:a cross-sectional study from elementary to secondary school[J]. Personality and Individual Differences, 2007, 42(3):441-451.
[17] Zafra A, Romero C, Ventura S. Multiple instance learning for classifying students in learning management systems[J]. Expert Systems With Applications, 2011, 38(12):15020-15031.
[18] Wang F, Chen L. A nonlinear state space model for identifying at-risk students in open online courses[C]//International Conference on Educational Data Mining. Raleigh:ERIC, 2016:527-532.
[19] He J, Bailey J, Rubinstein B, et al. Identifying at-risk students in massive open online courses[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Austin:AAAI, 2015:1749-1755.
[20] 周志华. 机器学习[M]. 北京:清华大学出版社, 2016.
[21] Hu C Y, Chen Y Q, Peng X H, et al. A novel feature incremental learning method for sensor-based activity recognition[J]. IEEE Transactions on Knowledge and Data Engineering, 2019, 31(6):1038-1050.
[22] Hou C P, Zhou Z H. One-pass learning with incremental and decremental features[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(11):2776-2792.
[23] Liu Y, Schiele B, Sun Q. Meta-aggregating networks for class-incremental learning[EB/OL]. arXiv:2010.05063. (2020-10-10)[2021-04-25]. https://arxiv.org/abs/2010.05063.
[24] Hu C Y, Chen Y Q, Hu L S, et al. A novel random forests based class incremental learning method for activity recognition[J]. Pattern Recognition, 2018, 78:277-290.
[25] Hou B, Zhang L, Zhou Z. Learning with feature evolvable streams[C]//Proceedings of the 31th Conference on Neural Information Processing Systems. Long Beach:MIT Press, 2017:1417-1427.
[26] He H B, Chen S, Li K, et al. Incremental learning from stream data[J]. IEEE Transactions on Neural Networks, 2011, 22(12):1901-1914.
[27] 余保玲, 虞松坤, 孙耀然, 等. 基于DeepPose和Faster RCNN的多目标人体骨骼节点检测算法[J]. 中国科学院大学学报, 2020, 37(6):828-834.
[28] 赖训飞, 梁旭文, 谢卓辰, 等. 基于实体嵌入和长短时记忆网络的入侵检测方法[J]. 中国科学院大学学报, 2020, 37(4):553-561.
[29] Aguiar E, Bhanpuri N, Lakkaraju H, et al. Student early warning:the United States, dssg[DB/OL]. (2018-08-22)[2021-04-25]. https://github.com/dssg/student-early-warning.

WILS：面向学业预警的非均衡增量式学习方法

A weighted incremental learning scheme for effective academic warning

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 12

编辑推荐

Metrics

本文评价

访问统计

联系我们

[1]	姚沐风, 昝露洋, 李柏鹏, 李庆亭, 陈正超. 基于CAR-Siamese网络的高分辨率遥感图像建筑物变化检测[J]. 中国科学院大学学报, 2023, 40(3): 380-387.
[2]	陈若男, 彭玲, 刘玉菲, 卫志超, 吕蓓茹, 陈德跃. 引入空间距离信息的城郊山区道路提取与应用[J]. 中国科学院大学学报, 2022, 39(5): 658-667.
[3]	古煜民, 阎福礼. 基于不同骨架UNet++网络的建筑物提取[J]. 中国科学院大学学报, 2022, 39(4): 512-523.
[4]	林大权, 范睿, 张良峰. 基于高维特征的图像对抗攻击算法[J]. 中国科学院大学学报, 2022, 39(3): 421-431.
[5]	赵敏钧, 赵亚伟, 赵雅捷, 罗刚. 一种新的基于深度学习的重叠关系联合抽取模型（英文）[J]. 中国科学院大学学报, 2022, 39(2): 240-251.
[6]	周龙, 王伟强, 吕科. FACR：一种快速且准确的车辆识别器[J]. 中国科学院大学学报, 2021, 38(1): 130-136.
[7]	李森, 彭玲, 胡媛, 池天河. 基于FD-RCF的高分辨率遥感影像耕地边缘检测[J]. 中国科学院大学学报, 2020, 37(4): 483-489.
[8]	卢凯旋, 李国清, 陈正超, 昝露洋, 李柏鹏, 高建威. 基于负样本多通道优化SSD网络的钢铁厂提取[J]. 中国科学院大学学报, 2020, 37(3): 352-359.
[9]	闫凯, 沈汀, 陈正超, 闫弘轩. 基于深度学习的SSD模型尾矿库自动提取[J]. 中国科学院大学学报, 2020, 37(3): 360-367.
[10]	宋旭鸣, 沈逸飞, 石远明. 基于深度学习的智能移动边缘网络缓存[J]. 中国科学院大学学报, 2020, 37(1): 128-135.
[11]	刘洋, 罗晨希, 罗铁坚. 和积网络的性质分析及其有效性验证算法[J]. 中国科学院大学学报, 2020, 37(1): 136-143.
[12]	杨宇, 张小雷, 雷军. 基于客货运量的新疆城市等级体系分析[J]. 中国科学院大学学报, 2010, 27(1): 27-35.