随着我国高等教育进入普及化阶段，人才培养质量引起社会各界广泛关注，提高人才培养质量是高等教育目标之一，学业成绩作为人才培养质量的重要体现，是高校人才培养过程中重点关注的内容。目前，高校学业成绩不佳的学生越来越多，如何有效地帮助这些学生提升学业成绩尤为迫切，大数据从技术层面提供了支撑。在数据驱动下，可以从学校日常产生的真实数据中开展学业成绩研究，使学校在提升学生学业成绩方面有的放矢，实现提高人才培养质量的目标。

研究者在高校从事信息化工作近二十年，一直负责学校校务数据的建设与管理，让“数据说话”，发挥数据价值，是研究者一直以来的工作追求。因此，选定了利用学校学生大数据开展学业成绩相关因素和学业成绩预测模型的研究，旨在通过对数据的挖掘分析，在数据中发现与学业成绩相关的因素，并基于这些因素构建学业成绩预测模型，使学校及时了解学生学业现状，为学生提供精准化服务，帮助学生提升学业成绩。研究对于提高学校人才培养质量，提升学校的现代化治理水平，具有重要的实践价值。

研究以案例大学为个案，在数据驱动下，基于学生大数据以大一、大二、大三三个年级的学生为研究对象，采用个案研究法开展研究。利用专家访谈法和大数据技术收集数据，通过类属分析法和数据挖掘法分析数据，在探讨国内外学生大数据、学业成绩相关因素以及学业成绩预测模型研究的基础上，重点解决了四个方面的问题：

一、构建了“高校学生学业成绩相关因素理论模型”。以社会认知理论和学生投入理论为依据，在数据驱动下，基于学生大数据构建了“高校学生学业成绩相关因素理论模型”。

二、开展了学业成绩相关因素的研究。首先，依据“高校学生学业成绩相关因素理论模型”构建了学业成绩相关因素体系框架，包括6个一级维度、8个二级指标和35个自变量，按照该框架从学生大数据中初步确定了学业成绩相关因素；然后，以大学学生管理系统、教务管理系统、校园一卡通系统、图书借阅系统、图书馆门禁系统、图书馆电子资源系统、校园网络管理系统、URL日志系统以及无线网络管理系统为学生大数据来源，对与学业成绩相关因素相关的数据进行整理，完成数据的采集、处理和存储。基于整理好的数据，通过Excel工具和编写计算机软件算法程序，完成了所有研究对象的35个自变量和因变量（学业成绩）的量化计算分析。为了构建精度高的模型，对35个自变量的共线性进行分析，对变量进行降维，采用主成分分析法（PCA）对量化的35个自变量提取数据主要特征变量，最终确定了17个变量为学业成绩相关因素。

三、开展了学业成绩预测模型研究。先对学生数据进行了描述性分析，了解学生数据的整体情况。再对学生学业成绩进行描述性分析，对人口统计学特征、在校状态，物理空间和信息空间学习生活行为三个维度的17个变量与学业成绩的相关性进行了初步分析。然后利用BP神经网络和卷积神经网络（LeNet、VGGNet和ResNet）两种方法构建了学业成绩预测模型，并对两种方法构建的模型进行比较，最终得出BP神经网络在模型精度（（Top1准确率为0.4944，Top2准确率为0.8301）和模型构建速度（105.9s）方面最优，作为本研究的学业成绩预测模型，为高校开展学业成绩预警与提升等管理工作提供支撑。采用SPSS计算了17个变量对学业成绩的贡献率，并从“个体因素”“行为因素”和“环境因素”三方面对17个变量进行了深层次的提炼归类，为高校以更聚焦的方式开展学业成绩提升工作提供依据。

最后，研究在学校、学院及个人层面提出了学业成绩提升策略。

本研究基于高校学生大数据形成了“高校学生学业成绩相关因素理论模型”、学业成绩相关因素、高校学业成绩预测模型等可用于实践的成果，在学业成绩相关因素的确定与学业成绩预测方面取得了突破，为高校人才培养质量的提升提供了有益的参考。

As higher education has entered the stage of popularization in China, the quality of talent training has attracted widespread attention from all walks of life. Improving the quality of talent training is one of the goals of higher education. Academic achievement as an important embodiment of talent training quality, is the focus of attention in the process of talent training in universities. At present, there are more and more students with poor academic achievement in colleges and universities. How to effectively help these students improve their academic achievement is particularly urgent. Big data provides support from the technical level. Driven by data, the research of academic achievement can be carried out from the real data generated by the university on a daily basis, so that the university can have a definite aim in improving the academic achievement of students and achieve the goal of improving the quality of talent training.

The researcher has been engaged in information work in colleges and universities for nearly 20 years, and has been responsible for the construction and management of college data. Letting the "data speak" and giving full play to the value of data is the researcher's long-term work. Therefore, the researcher selected the use of school students' big data to carry out the research on the related factors of academic achievement and the prediction model of academic achievement. Through the analysis of the data mining, the researcher found the factors related to academic achievement in the data, and built the academic achievement prediction model based on these factors. This enables the college to keep abreast of students' academic status, provides students with accurate services, helps the student improve the academic achievement. The research has important practical value for colleges and universities to improv the quality of talent training and enhance the level of modernization of college governance.

This research took the case university as a case, took freshmen, sophomores and juniors as the research objects based on student big data driven by data，and adopted the case study method to conduct research. Collected data by using the expert interview method and big data technology, and analyzed the data by the generic analysis and data mining methods. On the basis of discussing the big data of students at home and abroad, the related factors of academic achievement and the prediction model of academic achievement, the study focused on solving the problems in four aspects: 

First of all, constructed the theoretical model of college students' academic achievement related factors. Based on social cognition theory and the student engagement theory, the researcher constructed the theoretical model of college students' academic achievement related factors based on student big data driven by data.

Secondly, carried out the research on the academic achievement related factors. First，according to t the theoretical model of college students' academic achievement related factors, the academic achievement-related factor system framework was constructed, including 6 first-level dimensions, 8 second-level indicators and 35 independent variables. According to the system framework, factors related to academic achievement are preliminarily determined from student big data.  Then, this researcher sorted out the data related to academic achievement related factors by taking the student management system, educational administration management system, campus card system, library lending management system, library access control system, library electronic resource system, campus network management system, URL log system and wireless network management system as the source data and completed data collection, processing and storage. Based on the sorted data, completed the quantitative calculation and analysis of 35 independent variables and dependent variables (academic achievement) about all research objects by using Excel tools and writing computer software algorithm program. In order to build a high-precision model, analyzed the collinearity of 35 independent variables and reduced the dimension of variables. Finally, 17 variables were identified as academic achievement related factors by using principal component analysis (PCA) to extract the main characteristic variables from 35 quantified independent variables. 

Thirdly, carried out the research on the prediction model of academic achievement. At first, the descriptive analysis of student data is made on to understand the overall situation of student data. Then, the descriptive analysis of students' academic achievement was made on. The correlation between the 17 variables of three dimensions from demographic characteristics, school status, physical space and information space learning and living behavior and academic achievement was preliminarily analyzed. And then, two model methods of the BP neural network and convolutional neural network (LeNet, VGGNet and ResNet) were used to construct the academic achievement prediction model, and the models constructed by the two methods were compared. This paper finally concluded that the BP neural network was the best in terms of model accuracy (the accuracy rate of Top1 is 0.4944, and the accuracy of Top2 is 0.8301) and model construction speed (105.9s), and regarded it as the Prediction Model of College Students' Academic Achievement. It provided support for colleges to carry out management work such as academic achievement warning and improvement of academic achievements. The research has calculated the contribution rate of 17 variables to academic achievement using SPSS. The study deeply refines and categorizes 17 variables from "individual factors", "behavioral factors" and "environmental factors", providing a basis for universities to carry out academic achievement improvement work in a more focused way.

At last, this research proposed strategies to improve academic achievement for universities, colleges and individual levels.

Based on student big data, this paper formed the theoretical model of academic achievement related factors, academic achievement related factors, prediction models of academic achievement in colleges and universities that can be used in practice. A breakthrough has been made in the determination of academic achievement related factors and the prediction of academic achievement. It provides a useful reference for improving the quality of talent training for colleges and universities.