摘要
海外留学门槛增高导致入学竞争激烈,准确预测入学率对于申请者具有重要的参考意义。现有的基于数据驱动方法构建的入学预测模型通常只对定量历史数据进行分析,没有考虑考核指标中包含的专家定性知识,且预测结果不具有可解释性。为了解决以上问题,提出了一种新的基于证据推理(Evidence Reasoning,ER)和置信规则库(Belief Rule Base,BRB)的入学率预测方法。利用ER实现多指标有效融合来达到降低模型复杂度的目的;通过BRB构建入学预测模型;最后采用基于投影协方差矩阵自适应策略(P-CMA-ES)算法优化模型参数,从而提高模型精度。通过UCLA研究生数据集验证了模型的有效性,实验结果表明,上述方法可以准确的预测申请者的入学概率,为申请者成功入学提供了保障。
The increase in the threshold for studying abroad has led to fierce competition for admission,and accu-rate prediction of enrollment rate if of reference significance for applicants.Existing enrollment prediction models based on data-driven methods usually only analyze quantitative historical data,without considering the qualitative knowledge of experts contained in the assessment indicators,and the prediction results are not interpretable.In order to solve this problem,this paper proposes a new enrollment prediction method based on Evidence Reasoning(ER)and Belief Rule Base(BRB).First,utilizing ER to achieve effective fusion of multiple indicators to achieve the purpose of reducing model complexity;Second,using BRB to build an enrollment prediction model.Finally,the projection covar-iance matrix adaptive estimation strategy(P-CMA-ES)algorithm is adopted to optimize the model parameters,thereby improving the model accuracy.The effectiveness of the model is verified by the UCLA graduate student dataset.The experimental results show that the method can accurately predict the applicant's enrollment probability and provide a guarantee for the applicant's successful enrollment.
作者
陈伟伟
朱海龙
许冰
穆全起
CHEN Wei-wei;ZHUHai-long;XU Bing;MU Quan-qi(Department of Computer Science and Information Engineering,Harbin Normal University,Harbin Heilongjiang 150025,China)
出处
《计算机仿真》
北大核心
2023年第11期218-225,356,共9页
Computer Simulation
基金
中国博士后科学基金项目(2020M683736)
黑龙江省自然科学基金项目(LH2021F038)
黑龙江省大学生创新实践项目(202010231009,202110231024,202110231155)
哈尔滨师范大学博士科研启动金项目(XKB201905)
哈尔滨师范大学研究生质量培养提升工程项目(1504120015)
哈尔滨师范大学研究生学术创新项目(HSDSSCX2021-120,HSDSSCX2021-29)。
关键词
入学预测
组合爆炸
证据推理
置信规则库
投影协方差自适应进化策略
Graduate admission
Combinatorial explosion
Evidential reasoning
Belief rule-base
Projection covar-iance adaptive evolution strategy
