Analysis And Prediction Method of University Students' Behavior Based on Big Data

doi:10.21203/rs.3.rs-2230379/v1

Download PDF

Research Article

Analysis And Prediction Method of University Students' Behavior Based on Big Data

https://doi.org/10.21203/rs.3.rs-2230379/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Under the background of "Internet + Big Data", this paper takes big data as the basis of research, selects three dimensions of user basic information, learner behavior information, and learning attitude information, builds feature indicators, establishes a learner behavior model through genetic k-means clustering algorithm, constructs learner portrait, and judges their behavioral patterns, learning motivation, and further predict their academic passing rate by the plain Bayesian algorithm, to provide educators with a basis for educational decision-making.

With the deep integration of information technology and human life, the "Internet + Big Data" model is sweeping in, penetrating all walks of life, and the global data has shown explosive growth, which has caused unprecedented concern from academia, industry, government departments, and other organizations, and countries have introduced policies on Big Data one after another. China has elevated big data to a national strategy, proposed the implementation of the national big data strategy, and promoted the integration of big data, AI, and the real economy in depth in the context of the Internet. With the development of big data, the education industry is also facing new challenges and opportunities. Liu Qingfeng, chairman of KDDI, has proposed: "In the era of big data and artificial intelligence, promote the establishment of the National Development and Reform Commission's big data special 'basic education big data research and development and application demonstration project' to promote the integration of teaching and big data, and through the recording and processing of student data, accurately analyze each student's learning situation, truly aiming at students' personalized development and growth, realizing education according to their abilities, and promoting the model reform and ecological reconstruction of education with new technical support." Thus, the teaching management process of colleges and universities should be combined with the development of big data, through the mining, processing, and analysis of educational big data, so that teachers can better understand students' knowledge mastery and learning preferences, helping administrators to discover the shortcomings of the current teaching system, and truly promoting traditional education from "teacher-centered" to "student-centered". It helps administrators to identify the shortcomings of the current teaching system and truly promote the transformation of traditional education from "teacher-centered" to "student-centered", to teach students according to their abilities.

With the continuous development of information technology, technologies such as the Internet of Things, cloud computing, and big data [1][2] have been widely used, and the continuous construction of digital campuses and "smart campus" in colleges and universities, and the increasing number of campus management application systems and campus card service platforms, the data accumulated in the campus information environment are also gradually expanding. The traditional campus management concept and data analysis methods can no longer meet the increasing data processing demands. How to efficiently manage and share campus data, optimize student management by using big data mining and analysis, and provide clearer and more detailed data services for students' campus life through analysis results is one of the problems faced by the construction of campus service systems today [3]. The uneven construction time and different technical architectures of campus applications bring about problems such as complex data types and low data quality, while the continuous growth of data scale poses serious challenges to data storage and analysis. Based on the above problems, it is of good practical significance to build a campus big data analysis and service platform with comprehensive coverage, safety, and reliability by integrating various data sources such as teaching, management, office, learning, and life. At present, with the continuous improvement of the informatization level of each university, a shared data platform including data on the University campus one-card system, student achievement data, teaching attendance data, library borrowing data, access control data, network access log data, and other data has been gradually formed, which provides a platform for analyzing, processing and mining potential data laws in the campus big data environment, studying the characteristics of student behavior, analyzing student behavior laws and improving It provides a data foundation for analyzing, processing and mining potential data patterns in the campus big data environment, studying the characteristics of student behavior, analyzing student behavior patterns, and improving school management decision-making capabilities. In addition to contributing to the improvement of school management and student service quality, the analysis and prediction of massive teaching management data through the application of big data technology [4][5][6] can also promote the progress of teaching mode and realize high-quality and personalized teaching. Big data can comprehensively record students' growth records and conduct scientific analysis, allowing students to understand themselves better. Through the big data platform, we collect timely behavioral data on students' campus classes, meals, book borrowing, dormitories, and physical exercise, allowing school leaders, faculty, academic staff, and teachers to grasp every student's school situation in all aspects. It helps instructors understand students' performance and how hard they are working in a school, and helps them improve their academic performance. To sum up, big data will play a very obvious role in the reform of school education, and it has obvious advantages in improving the quality of University management and teaching, as well as improving the means of student education evaluation. This paper relies on the digital campus platform of the school to research and develop a student behavior analysis and prediction service platform based on campus big data. By designing a student behavior description index system and adopting the relevant methods of big data, we analyze students' learning and consumption behaviors at school and provide a basis for objective and comprehensive evaluation of students, which is of good practical significance to the improvement of school education and teaching quality.

In the context of big data, universities should establish certain data awareness and data thinking, open educational platforms, and create "educational databases". To enhance the application value of big data in education, Zhu Lin establish a professional data analysis team to improve the management efficiency students' work [7]; Ref[8] points out that the concept of education should be changed, the development of students should be the main focus, the teaching structure should be readjusted, the information transfer method should be reorganized, and a real "learning community" should be formed; Ref[9] uses data-driven technology to generate personalized learning maps for students to achieve the purpose of precise education and precise learning. The application of educational big data in the teaching reform of universities has been common. Ref[4] improves students' job matching through big data analysis to dovetail with national industrial development plans; Ref[10] proposes to better grasp students' learning status with the help of big data to ensure the effectiveness and relevance of teaching; Ref[11] propose that collaborative education projects can be carried out through the collection and processing of big data, combined with enterprises to implement the combination of industry and education. In response to the series of problems in traditional education, the use of data mining algorithms and technical means to achieve targeted and personalized teaching has gradually become a hot issue of current research. Ref[12] analyzed the learning needs and learning styles of different students combined with big data technology to carry out personalized teaching; Ref[13] implemented a digital campus through big data analysis and established a teaching effect or teaching feedback database to provide a basis for teaching evaluation; Xia Yang designed and optimized a classroom interactive feedback system, which can present teacher-student ground interaction in real-time and facilitate teachers' post-class The interactive feedback system was designed and optimized by Ref[14], which can present teacher-student interactions in real-time and facilitate teachers to view them after class. The collected and mined educational big data are analyzed to establish learning and teaching evaluation indexes including students and teachers to improve the traditional teaching system and truly personalize teaching and learning according to the student's needs. Ref[15] predicts students' learning paths and plans students' learning strategies by collecting and visualizing student data; Ref[16] combines teachers' teaching experience and students' learning data as a way to improve students' learning trajectories; Ref[17] combines the private customized education provided in the smart classroom to achieve multidimensional communication and interaction and promote data-based teaching decisions. Foreign research on personalized instruction is aided by various types of data mining algorithms for analysis. Ref[18] provides targeted and intensive courses at the lowest possible cost by collecting and recording longitudinal student data from the Ministry of Public Education and applying classification methods of decision trees and support vector machines; Ref[19] provides targeted and intensive courses by recording student activities in an e-learning platform, performing statistical and association rule techniques using a big data framework and evaluating them to identify trends and deficiencies in students' use of this learning management system; Ref[20] by proposing a conceptual model that captures higher education system's student students, teachers, administrators, and other information that can be more successfully implemented in higher education for analysis. Therefore, the use of educational big data for teaching and learning analysis and reform is a hot topic of research at home and abroad, through the specific use of data mining algorithms for educational big data The analysis and visualization of data, the clustering analysis of student data using a k-means clustering algorithm, so that personalized teaching models can be adopted for different characteristics of each cluster class, the use of plain Bayesian prediction of academic performance, optimization of teachers' teaching content, and the provision of technical and algorithmic support for teaching reform under educational big data to improve the accuracy of teaching.

3.1 Selection of data set

Analysis and prediction of student learning behavior based on educational big data was selected from the public dataset in UCF-101 and HMDB51 dataset, which has a total of 641,139 data, including 465332 data from University records and 175817 data from college records, containing data from 16 courses in three semesters, fall and spring, for the academic years 2018 to 2021, a collection of data generated from courses and learner activities in courses, mainly 465332 data from University records were selected from the above dataset for the study.

3.2 Construction of feature indicators

The data provided by the university in the big data set were mainly selected for analysis, mainly in three dimensions: basic user information, learner behavior information, and learning attitude information.

(1) Basic information of users: education (Loe_DI), gender (gender), age (age).

(2) Learner behavior information: learning start time (start_time_DI), learning end time (last_event_DI), number of course interactions (nevents), number of days of access to learning (ndays_act), number of video plays (nplay_video), number of chapters of learning (nchapters), and The number of posts (nforum_posts), where the difference between the end time of learning and the beginning time of learning is the characteristic indicator of the length of learning, the number of course interactions refers to the number of times learners log in to their learning accounts, and the number of posts refers to the number of times learners speak in discussion forums, forums, etc. to raise questions for course communication.

(3) Learning attitude information: learning attitude diffuse (viewed), learning attitude positive (explored), where learning attitude diffuse means that the learner has registered for the course and carried out a certain degree of learning, learning to attitude positive means that the learner has not only registered for the course, and carried out more than 50% of the course content learning. The above characteristic information indicators were established as shown in Table 1.

Table 1

Selection of characteristic indicators
dimensions	indicators
Basic user information	Education, gender, age
Learner behavior information	Length of study (end time of the study - start time of the study), number of course interactions, number of days of study visits, number of video plays, number of study chapters, number of forum posts
Learner Attitude Information	Scattered learning attitude, positive learning attitude

3.3 Data Cleaning and Processing

The selected dataset must be cleaned and pre-processed because of data problems such as null, noise, error values, and differences in the magnitudes of the selected feature indicators. This is done mainly in two ways:

3.3.1 Differences in data quality

For the collected data sets, data that do not affect the actual results, data that are missing some characteristic values, data that have obvious errors, etc. should be eliminated, and data should be screened mainly from the following aspects:

(1) Filter out learner data where any of the indicators involved in the dataset are empty.

(2) Filter out data on learners who are registered for a course but have not attended any of the courses.

(3) Filtering out data information such as the number of instructor comments and replies mixed in the dataset.

(4) filtering out the data information of learners whose learning end time is earlier than the learning start time;

(5) filtering out the data information of learners who have obvious errors and exist outside of the group.

3.3.2 Differences in data outline

Because of the differences in the feature indicators of different dimensions in the selected dataset, the data types and magnitudes are different among each attribute, which will bring large errors if directly used for the operation and analysis of mathematical models and data algorithms, and thus affect the final results. Therefore, the data set needs to be normalized to focus its data range in the (0, 1) interval, or (-1,1) interval, etc., and then run and analyze the data.

4.1 Basic idea of the algorithm

Clustering algorithms belong to the unsupervised type of big data algorithms, and the main implementation process is to move the dataset from unlabeled to labeled. Since the traditional k-means clustering algorithm has a random selection of the initial clustering centers, it further has an impact on the performance of the clustering algorithm. Based on the characteristics and metrics of the data in most of the selected datasets, combined with the excellent global search capability in the genetic algorithm, the genetic algorithm-based k-means optimized clustering algorithm is therefore used for the implementation of student portraits. The main algorithm steps are shown in Table 2.

Step	Genetic k-means algorithm
Step1	Parameter settings: number of data set samples S, number of clusters K, population size P, the maximum number of iterations T, crossover probability Pc, variance probability Pm
Step2	Population initialization: K points are randomly selected from the sample as the initial clustering centers and encoded, repeated P times, to generate the initial population
Step3	Calculating the fitness of individuals in a population
Step4	Selection operations for populations based on fitness values
Step5	Crossover operation of selected individuals
Step6	Mutational manipulation of individuals after crossover to generate new populations
Step7	The k-means operation is performed on each individual in the new population generated in the previous step
Step8	Generating a new generation of populations
Step9	Determine if the end condition is satisfied. If it is satisfied, output the result; otherwise, go to Step3

The main idea of the genetic k-means algorithm is divided into two parts: firstly, the initial population is selected based on the data set, and the genetic algorithm is run; secondly, the new population generated in the genetic algorithm is analyzed by the k-means clustering algorithm.

(1) Establish sample evaluation index matrix

$$A=\left[\begin{array}{cccc}{Y}_{11}& {Y}_{12}& \cdots & {Y}_{1n}\\ {Y}_{21}& {Y}_{22}& \cdots & {Y}_{2n}\\ ⋮& \cdots & \ddots & ⋮\\ {Y}_{n1}& {Y}_{n2}& \cdots & {Y}_{mn}\end{array}\right]$$

(2) Dimensionless quantification of evaluation matrix

$${Z}_{ij}=\left\{\begin{array}{c}{Y}_{ij}/{Y}_{jmax}\mid {Y}_{jmax}=max\left({Y}_{ij}\right)\\ {Y}_{jmin}/{Y}_{ij}\mid {Y}_{jmin}=min\left({Y}_{ij}\right)\end{array}\right.$$

(3) Establish standardized decision matrix R. where R_ij=W_j*Z_ij

(4) Calculate positive and negative ideal solutions

$$\left\{\begin{array}{c}{X}^{+}=\left[\left(max{R}_{ij}/j\right)/i\right]=\left[{X}_{1}^{+},{X}_{2}^{+}\cdots {X}_{n}^{+}\right]\\ {X}^{-}=\left[\left(max{R}_{ij}/j\right)/i\right]=\left[{X}_{1}^{-},{X}_{2}^{-}\cdots {X}_{n}^{-}\right]\end{array}\right.$$

(5) Calculate the distance between varieties and positive and negative ideal solutions

$$\left\{\begin{array}{c}{S}_{i}^{+}=\sqrt{\sum _{i=1}^{n} {\left({R}_{ij}-{X}_{i}^{+}\right)}^{2}}\\ {S}_{i}^{-}=\sqrt{\sum _{i=1}^{n} {\left({R}_{ij}-{X}_{i}^{-}\right)}^{2}}\end{array}\right.$$

The main problems of the method are as follows.

(1) Performing genetic coding.

(2) Select and construct a suitable fitness function based on the fitness of the gene string for clustering. The higher the result of the fitness function, the better the clustering performance, and vice versa.

(3) When performing genetic algorithm operations, the values of the genetic parameters involved are further determined. According to the selected dataset characteristics and algorithm performance, the adoption of floating-point genetic coding based on the clustering center can reduce the genetic k-means algorithm main idea is divided into two parts: firstly, the initial population is selected according to the dataset and the genetic algorithm is run; secondly, the new population generated in the genetic algorithm is analyzed by the k-means clustering algorithm.

According to the selected dataset characteristics and algorithm performance, a floating-point genetic encoding based on the cluster center can reduce the number of operations and simplify the operation process while adapting to the dataset characteristics. In determining the fitness function, the criterion function in the k-means algorithm is used to represent: f = 1 ⁄(1 + J), and the smaller the value of J, the larger the value of the fitness function. The roulette wheel method is used for the setting of the selection operator, which randomly sets the intersection of the arithmetic crossover to protect the optimal individuals and improve the convergence of the algorithm. Therefore, the main parameters of the genetic k-means algorithm are set as follows: the number of samples of the cleaned data set is 25216, the number of clusters is set to K = 3, the population size P = 100, the number of iterations T = 200, the crossover probability Pc = 0.9, and the variation probability Pm = 0.01

4.2 Algorithm-specific operation process

4.2.1 Encode

The feature attributes in the student dataset are denoted in order as j, where j = 1, 2, ..., 11, the attributes denoted in order are education, gender, age, study duration, number of course interactions, number of days of study visits, number of video plays, number of study chapters, and number of postings. The number of clusters K = 3 classes and the number of data sets i (i = 1, 2,..., n), where n = 25216, can form a data matrix of i × j.

$$\left[\begin{array}{ccccc}{x}_{11}& {x}_{12}& \dots & {x}_{10}& {x}_{11}\\ {x}_{i1}& {x}_{i2}& \dots & {x}_{i10}& {x}_{i11}\\ \dots & \dots & \dots & \dots & \dots \\ {x}_{n1}& {x}_{n2}& \dots & {x}_{n10}& {x}_{n11}\end{array}\right]$$

First, each data object is normalized to [0, 1] using Eq. 1,

$${x}_{ij}^{{\prime }}=\frac{\left|{x}_{ij}-min\right|}{max-min}$$

Then the floating-point encoding of chromosomes and the dimensionality of this learner dataset is 11-dimensional, so the choice is to encode the cluster centers, that is, K j-dimensional vectors indicating the cluster centers are connected, the length of chromosomes formed after encoding, and the gene positions take values in the range [0, 1], and randomly selected until the final formation of chromosomes with the number of individuals of the initial population p(0) of 100.

4.2.2 Genetic manipulation

Roulette wheel selection method based on the initial population p(0):

$${p}_{si}={f}_{i}/\sum _{j=1}^{M} {f}_{i}$$

where the fitness of chromosome i is fi, i.e., the higher the fitness the greater the chance of the individual being selected; secondly the selected chromosomes were subjected to the arithmetic crossover of gene positions with a probability of Pc = 0.9, and finally a mutation operation with a probability of 0.01 was done on the chromosomes.

4.2.3 Calculating the degree of adaptation

According to the fitness function of chromosomes constructed in the text.

$${f}_{i}=\frac{1}{1+J}$$

$$J=\sum _{i=1}^{k} \sum _{x\in {c}_{i}} {w}_{i}\text{d}\text{i}\text{s}\text{t}{\left({c}_{i},x\right)}^{2}$$

The chromosomes undergoing selection, crossover, and mutation operations are calculated for their fitness, and those with the lowest fitness are eliminated and replaced with chromosomes with the highest fitness in the initial population p(0) to form the new generation population p(1).

4.2.4 Result Output

Keep repeating the above steps until the iteration termination number T = 200 set in this paper is reached, the algorithm stops, the final population p(T-1) is output, the chromosome fitness in this population is calculated, and the chromosome with the highest fitness is output, then the three five-dimensional data points represented by this chromosome are the optimal clustering centers of the weighted k-means algorithm. Finally, the data is clustered with this clustering center to get the cluster ci(i = 1,2,3). The genetic k-means clustering algorithm was run on the learner dataset in the MATLAB 2014 experimental environment. After clustering, the data of cluster class 1 is 9966 items, cluster class 2 is 13648 items, and cluster class 3 is 1602 items. By analyzing the characteristics of the above three cluster classes, it is easy to find that although the gender distribution in each cluster class varies greatly, in the overall filtered and processed dataset, 46% are female and 54% are male, and the difference between them is small. In all three clusters, the year of birth is concentrated between 1990–2000, accounting for more than half of the total. Regardless of the cluster, most of the learners' learning time is concentrated within 1000 days, with less than 1000 course interactions, the number of learning visits is generally within 30 days, the number of video plays is also more within 1000, and the number of learning chapters is concentrated within 5 lessons, and they hardly post in the forum, and because the online courses are less binding for learners, so 85% of the learners have a scattered learning attitude, both registering for the course and not learning more than 50% of the content of the course, and only 15% of the learners have a more positive learning attitude. The mean values of learner characteristic indicators of different cluster categories are shown in Table 3

Table 3

Mean values of characteristic indicators for different clusters of learners
Cluster	Cluster 1	Cluster 2	Cluster 3
Number of people	9966	13648	1602
The average length of study	572.30	571.35	1388.84
Mean number of interactions	406.17	761.07	3790.55
The average number of study visit days	5.14	8.09	26.86
The average number of video plays	67.76	132.31	619.59
The average number of study chapters	2.69	3.82	10.30
The average number of forum posts	0.01	0.03	0.04
Learning attitude mean	0.05	0.15	0.79
Number of competition certificates obtained	197	997	726

According to the mean values of learner characteristics indicators provided in Table 3, it can be seen that the number of learning interactions, days of learning visits, number of video plays, number of learning chapters, number of posts, and content of learning courses in Cluster 1 are the lowest among the three clusters, and the percentage of competition certificate recipients is less than 2%, so Cluster 1 can be clustered as a casual learner; various characteristics such as learning hours, number of interactions and days of learning visits in Cluster 2 has improved compared to Cluster 1, but are still lower than Cluster 3, and the percentage of competition certificate recipients is 7.3%. The indicators in Cluster 2 have improved compared to Cluster 1, but they are still lower than in Cluster 3, and the percentage of learners with competition certificates is 7.3%, so Cluster 1 can be clustered as average learners; the indicators in Cluster 3 are much higher than the first two clusters, and the percentage of learners with competition certificates can reach 45%, so it can be clustered as active learners.

Through genetic K-means clustering analysis of the learner dataset, three different cluster classes are obtained and three different learner profiles are constructed, namely, the scattered learner, the average learner, and the active learner. To further realize the accurate pushing of teaching contents, meet the personalized learning of learners, and improve the academic passing rate, in addition to the construction of learner profiles, it is also necessary to conduct academic The classification algorithm is comparable to the clustering algorithm. The classification algorithm is a supervised learning algorithm compared to the clustering algorithm, i.e., the data categories in the known dataset are used to test the unknown dataset categories. By analyzing the characteristics of the learner dataset and combining the classification performance of the classification algorithm, the plain Bayesian classification algorithm is used to analyze and predict the academic pass rate of the learners. The data in 95% of the three cluster classes were selected as the training set data of the plain Bayesian algorithm, and the remaining 5% of the data where the test set. The MATLAB run of the algorithm was used to predict the probability of obtaining a certificate in the test set data and compared with the real data, and the prediction accuracy in cluster class 1 was 99.6%, 97.37% in cluster class 2, and 78.05% in cluster class 3. This shows that the accuracy of using the plain Bayesian algorithm to predict obtaining certificates is high under different conditions of learner characteristics. Through the establishment of the learner learning behavior analysis and prediction model, we can realize the "visualization" of data, which can help teachers pinpoint students' weaknesses, customize personalized learning programs according to students' learning interests and learning abilities, and realize differentiated teaching and tailor-made teaching. The main implementation process is as follows: Precise design before class: the preparation process before class is the foundation of the whole teaching process, teachers should understand students' basic knowledge level, learning habits, course content tendency, subject background, cognitive ability, life experience, and other dimensions in advance, and have a comprehensive and detailed understanding of them, to design targeted pre-study content, push pre-study materials and assign pre-study exercises, to monitor and check students' To monitor and check students' progress and mastery, we flexibly select teaching methods and resources according to students' realistic needs and teaching objectives, and provide the basis for subsequent classroom teaching. Precise teaching in class: Based on the precise analysis of students before class, the traditional teaching mode should be changed from "teacher-centered" to "student-centered" in classroom teaching, through a variety of interactive forms such as flipped classrooms, group discussion, and individual presentation. The classroom atmosphere is enlivened and students' participation and motivation are increased. At the end of the class, the content of the course is sublimated and ideological and political education is implemented to guide students to form correct values and ideological and political identity and to lead them to practice this spirit and values in their behavior and life. After-class precision counseling: for the analysis of the two processes before and during the class, teachers through the student's mastery of the class content and the ability to digest the knowledge points, targeted assignments, pushing different teaching resources, statistics on the wrong questions, differentiated counseling on the weak links of students' mastery of knowledge, early warning of individual poor students' learning performance and intervention in their learning styles. Teachers can also use education data to reflect on and improve teaching links, teaching contents, teaching methods, and teaching evaluation, and through adjustment, design a teaching process more suitable for students, better realize personalized teaching and achieve teaching according to their abilities.

By using the dataset provided by University in the big data set as the main data set, analyzing the characteristics of their learning groups through three dimensions of basic user information, learner behavior information and learning attitude information, applying the index selection method, from which we select education, gender, age, learning start time, learning end time, number of course interactions, number of days of learning visits, number of video plays, number of learning chapters, number of postings, the We use the k-means clustering algorithm in data mining to establish a student behavior model, build a student portrait, and then judge their behavior patterns, evaluate their learning motivation and learning interest, and classify them with the plain Bayesian classification algorithm, calculate the academic certificate pass rates of different clusters, predict their academic performance and knowledge tendency, and predict their academic performance and knowledge tendency by comparing with the original data Therefore, in the subsequent learning process, teachers can analyze individual differences according to this learning behavior analysis and prediction model, analyze individual differences according to the influencing factors on academic passing rate, make certain degree of learning interventions, push learning resources and contents that meet learners' characteristics In addition, teachers can change teaching methods, optimize teaching contents, standardize teaching process, and personalize teaching, which will lead to the transformation of traditional education from "teacher-centered" to "student-centered". The openness of the dataset provides the basis for big data analysis and processing in the education field. Based on the dimensions and characteristics of the selected dataset, we need to continue to optimize the relevant research algorithms in future education data analysis and continue to dig and study the dataset more deeply.

Declarations

Ethical Approval

Not applicable

Competing interests

No, I declare that the authors have no competing interests as defined by Springer, or other interests that might be perceived to influence the results and/or discussion reported in this paper.

Funding

The paper didn’t receive any funding.

Zhu Lin (2019) Student management reform based on education big data. J Yanbian Inst Educ 33(05):174–176
Chen Xiang Z, Liqun G(2019) Xiangyun. Thinking on mathematics classroom teaching in the context of big data era.Modern Agriculture, (11):110–111
Wu, Fenfen(2020) Zhang Zhen. Data driven research on precision teaching mode of higher vocational education based on smart classroom.Information Technology Exploration, (7):134–135
Xiong Xiaochun (2016) Re interpretation and re cognition of China's higher education teaching management in the context of big data.China Adult Education, (15):52–54
Jia Shuyi (2020) The application of big data in fine teaching in colleges and universities.China Rural Education, (01):12–13
Wang Yanli Z, Min Y, Guoqiang (2020) Preliminary Exploration on the Construction of Data Science and Big Data Technology under the Background of New Engineering.Shaanxi Education (Higher Education), (02):44–45
Lin Zhu (2020) The application of big data technology in mathematics teaching.Wireless Internet Technology, (04):148–149
Zhang H(2019) Opportunities and challenges of college teaching management in the age of big data.Journal of Qingdao Administrative College, CPC Qingdao Municipal Party School, (01):58–62
Wu Fenfen (2020) Design and implementation of classroom interactive feedback system based on learning big data.Expert Forum, October 11,
Jia Shuyi (2020) Research on personalized learning mode based on big data technology.Communication Design and Application, :66–67
Wang Y(2016) Analysis of personalized learning mode based on education big data.Teaching and Management, :101–103
Mou Lijiao (2020) Reform and practice of smart classroom teaching of ideological and political courses in colleges and universities in the age of big data - taking the outline course of modern Chinese history as an example.Journal of Chongqing University of Science and Technology (Social Science Edition), (3):115–120
Lucy C, Sorensen. “Big Data” in Educational Administration: An Application for Predicting School Dropout Riskã€‚Educational Administration Quarterly,2019, 55(3):404–446
Magdalena Cantabellaã€‚ Analysis of student behavior in learning management systems through a Big Data framework. Future Generation Computer Systems,2019(90):262–272
Mohsen Attaran. Opportunities and challenges for big data analytics in US higher education: A conceptual model for implementation.Industry and Higher Education,2018, 32(3):169–182
Candra Wijaya;Manshuruddin (2020) Zulham Effendi; "Character Building Through School Culture Development in The Senior High School of Panca Budi Medan". JOURNAL OF HUMANITIES AND SOCIAL SCIENCES
Yanxia Zhao; Wei Ren (2020) ; Zheng Li; "Prediction of English Scores of College Students Based on Multi-source Data Fusionand Social Behavior Analysis", REV. D'INTELLIGENCE ARTIF.,
Dingde Jiang; Liuwei Huo (2020) Houbing Song; "Rethinking Behaviors and Activities of Base Stations in Mobile Cellular Networks Based on Big Data Analysis". IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING. (IF: 4)
Lili Lv(2021) ; Yijun Xu; "Characteristics of Aerobics Teaching Based on Big Data Analysis", JOURNAL OF PHYSICS: CONFERENCE SERIES,
Wolfgang Wiedermann, Keith C (2021) Herman; Wendy Reinke; Alexander von Eye. "Configural Frequency Trees", DEVELOPMENT AND PSYCHOPATHOLOGY

Authors' contributions

All authors have the equal contribution.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Analysis And Prediction Method of University Students' Behavior Based on Big Data

Status:

Version 1

Abstract

1. Introduction

2. Related Work

3. Selection and preliminary analysis of student behavior characteristics indicators

3.1 Selection of data set

3.2 Construction of feature indicators

3.3 Data Cleaning and Processing

3.3.1 Differences in data quality

3.3.2 Differences in data outline

4. Construction and Analysis of Student Portraits Based on Genetic k-means Algorithm

4.1 Basic idea of the algorithm

4.2 Algorithm-specific operation process

4.2.1 Encode

4.2.2 Genetic manipulation

4.2.3 Calculating the degree of adaptation

4.2.4 Result Output

5. Analysis of students' learning behavior and academic prediction based on plain Bayesian algorithm

6. Conclusions

Declarations

Declarations

Funding

References

Unsectioned Paragraphs

Additional Declarations

Status:

Version 1