DEEP LEARNING-BASED PREDICTION OF ACADEMIC PERFORMANCE OF STUDENTS IN HIGHER EDUCATION INFLUENCED BY INTERNET USAGE

Abstract

ABSTRACT The method for the development of three categories of parameters were used to make the predictions: faculty, department, and midterm exam grades. These data-driven studies are crucial for developing a framework for learning analysis in higher education and for influencing decision-making. Lastly, this study identifies the best machine learning techniques and contributes to the early identification of students who are at high risk of failing. Four thousand students' actual Internet usage data was used to extract, compute, and normalize a set of features, such as online duration, Internet traffic volume, and connection frequency. Academic performance is positively connected with features of Internet connection frequency, while academic performance is negatively connected with features of Internet traffic volume. Predicting students at risk and students dropping out are two fundamental problems that are understood and resolved through the use of machine learning (ML) techniques. The online learning platforms and student databases from colleges and universities are the two types of datasets used in the majority of studies. It has been established that machine learning techniques are crucial for anticipating at-risk students and dropout rates, which enhances student performance.

Specification

Description:DEEP LEARNING-BASED PREDICTION OF ACADEMIC PERFORMANCE OF STUDENTS IN HIGHER EDUCATION INFLUENCED BY INTERNET USAGE
Technical Field
[0001] The embodiments herein generally relate to a method for deep learning-based prediction of academic performance of students in higher education influenced by internet usage.
Description of the Related Art
[0002] In recent years, there has been a lot of interest in the use of data mining techniques in the field of education. The process of finding data is called data mining (DM). It is the study of using big data to extract new, potentially valuable information or significant findings. It also uses various classification algorithms to extract new patterns and trends from big datasets. The Internet has been shown to improve educational attainment in a study. Additionally, a wealth of learning resources that support students' academic success have emerged with the rise of MOOCs and e-learning platforms. However, students' academic progress can also be adversely affected by the Internet. The Internet's widespread use has triggered a problem with self-control: problematic Internet use. There is evidence linking internet usage habits to academic achievement. Based on the needs of students, the diverse range of research has uncovered and imposed new opportunities for technologically enhanced learning systems. The EDM's cutting-edge approaches and implementation strategies are essential to improving the educational setting. For instance, by assessing both the educational environment and machine learning methodologies, the EDM plays a crucial role in comprehending the student learning environment.
[0003] EDM is the application of DM techniques to educational data, including student information, academic records, test scores, class participation, and the frequency of questions. EDM is now a useful tool for predicting academic success, uncovering hidden patterns in educational data, and enhancing the teaching and learning environment. The use of EDM has added a new dimension to learning analytics. Learning analytics encompasses the many facets of gathering student data collectively, analyzing and interpreting it to gain a deeper understanding of the learning environment, and identifying the best student/teacher performance. Numerous intriguing connections between online behavioral factors and academic achievement were found through analysis of these data. For instance, students' academic self-efficacy can be enhanced and their academic performance positively correlated when they use the Internet for general or educational purposes.
[0004] Students' academic success can be accurately predicted by behaviors in a learning management system. Together, smartphones and the Internet cause people to spend more time browsing the web than studying, which has a detrimental impact on academic performance. The environment in which modern educational institutions operate is complex and fiercely competitive. Therefore, some of the challenges that most universities today face are analyzing performance, offering high-quality education, developing strategies for evaluating the students' performance, and identifying future needs. Universities use student intervention plans to help students who are struggling academically. Predicting student performance at the entry level and in later phases aids universities in efficiently creating and modifying intervention plans, which benefit both management and teachers. The information gathered about educational procedures presents fresh chances to enhance the educational process and maximize users' engagement with technology.
SUMMARY
[0001] In view of the foregoing, an embodiment herein provides a method for deep learning-based prediction of academic performance of students in higher education influenced by internet usage. In some embodiments, wherein the predicting students' academic performance at the conclusion of a four-year program is the first component. The second is to look at students' growth and combine it with forecasted outcomes. He separated the pupils into two groups: those with low achievement and those with high achievement. He has discovered that in order to provide timely warnings, support underperforming students, and provide guidance and opportunities to high-performing students, educators should concentrate on a limited number of courses that show exceptionally good or poor performance. Cruz-Jesus et al. used 16 demographic factors, including age, gender, number of classes taken, internet access, computer ownership, and class attendance, to predict students' academic performance. The first part is predicting students' academic performance at the end of a four-year program. The second is to examine students' progress and integrate it with anticipated results. He divided the students into two groups: those who performed poorly and those who performed well. He has found that teachers should focus on a small number of courses that exhibit exceptionally good or poor performance in order to give timely warnings, support underperforming students, and give high-performing students guidance and opportunities. Cruz-Jesus et al. predicted students' academic performance using 16 demographic factors, such as age, gender, number of classes taken, internet access, computer ownership, and attendance at class.
[0002] In some embodiments, wherein a machine learning-based model to pinpoint the main elements influencing schools' academic achievement and ascertain how these elements relate to one another. He came to the conclusion that the regression trees demonstrated that the most significant variables linked to improved performance were gender proportions, school size, competition, class size, and parental pressure. Furthermore, the results of the random forest algorithm showed that the percentage of girls and the size of the school had a significant effect on the model's predictive accuracy. Data on Internet usage was gathered from the Internet access system of a university. Students' use of the Internet was documented, including how they accessed campus resources using their accounts. The dataset includes information on terminal device type, Internet download and upload volumes, online and offline times, and more.
[0003] In some embodiments, wherein these characteristics were primarily associated with psychomotor skills, including performance in the course and prior to it, student involvement, student demographics, including gender, high school achievement, and self-control. However, student motivation, habits, lack of advancement, social and financial problems, and career changes were the main factors influencing the dropout rates. The detailed times of login and logout are represented by the online and offline times, which are always generated in timestamp format; the data from the Internet to the user terminal is represented by the Internet download volume, and the data from the user terminal to the Internet is indicated by the Internet upload volume. The Internet access system calculates these volume columns from network packages, which can be explained by the unit byte.
[0004] These and other aspects of the embodiments herein will be better appreciated and understood when considered in conjunction with the following description and the accompanying drawings. It should be understood, however, that the following descriptions, while indicating preferred embodiments and numerous specific details thereof, are given by way of illustration and not of limitation. Many changes and modifications may be made within the scope of the embodiments herein without departing from the spirit thereof, and the embodiments herein include all such modifications.

BRIEF DESCRIPTION OF THE DRAWINGS
[0001] The embodiments herein will be better understood from the following detailed description with reference to the drawings, in which:
[0002] FIG. 1 illustrates a method for deep learning-based prediction of academic performance of students in higher education influenced by internet usage according to an embodiment herein; and
[0003] FIG. 2 illustrates a method of the workflow of the designed model according to an embodiment herein.
DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS
[0001] The embodiments herein and the various features and advantageous details thereof are explained more fully with reference to the non-limiting embodiments that are illustrated in the accompanying drawings and detailed in the following description. Descriptions of well-known components and processing techniques are omitted so as to not unnecessarily obscure the embodiments herein. The examples used herein are intended merely to facilitate an understanding of ways in which the embodiments herein may be practiced and to further enable those of skill in the art to practice the embodiments herein. Accordingly, the examples should not be construed as limiting the scope of the embodiments herein.
[0002] FIG. 1 illustrates a method for deep learning-based prediction of academic performance of students in higher education influenced by internet usage according to an embodiment herein. In some embodiments, all available student data is routinely stored in electronic format by educational institutions. Databases are used to store and process data. These data can range in volume and type from student demographics to academic performance. The Student Information System (SIS), where all student records are kept at a State University in Turkey, provided the data for this study. We selected 1000 students at random from each of the two sub-datasets to ensure that the sample size was balanced. the sample students' Internet usage statistics. After removing the National Day from the dataset, there were no holidays for a total of 80 days, from October 8, 2016, to December 26, 2016. As a result, the majority of students were expected to live on campus and maintain ongoing records of their Internet usage. Between the two datasets, we were able to obtain over 20 million records from 4000 sample students. To find a list of studies that can be used for additional assessment, a search is conducted. A bibliography tool called Mendeley handles the studies' bibliography management. The studies that fully meet the inclusion criteria are included in these bibliographies. The 78 papers that were produced after the inclusion and exclusion criteria were successfully applied.
[0003] In some embodiments, the final exam is given at the end of the semester, and the midterm is typically given in the middle of the academic year. The time between the midterm and final exams is roughly nine weeks. In other words, students who are at risk of failing because of the final exam predictions have two and a half months to take corrective action. Stated differently, the degree to which a student's performance during the middle of the semester influences his performance at the end of the semester was examined. The raw data, which also shows the frequency of Internet connections with various terminals throughout the day, can be used to determine the preference of mobile or PC devices. The average amounts of Internet data used daily are also computed, in addition to the characteristics of online time and connection frequency. Predicting student performance has great advantages for raising retention rates, managing enrollment and alumni effectively, improving targeted marketing, and enhancing the overall efficacy of educational institutions. Students who are at risk of not graduating are assisted by school intervention programs. The precise and prompt identification and prioritization of the students in need of support is the foundation for the success of such programs. In this section, at-risk student performance using machine learning techniques is documented through a chronological review of published literature from 2009 to 2021.
[0004] In some embodiments, at this stage, it is decided which data source will be used, which data features will be utilized, and whether the gathered data is appropriate for the intended use. Reducing the number of variables used to predict a specific outcome is known as feature selection. The objective is to make the model easier to understand, simplify it, make algorithms more computationally efficient, and prevent overfitting. The relationship between the aforementioned behavioral traits of students and their academic performance is measured using Spearman's correlation coefficient. Lastly, three popular machine learning predictive algorithms for classification-decision tree (DT), neural network (NN), and support vector machine (SVM)-are used to assess the substantive validity of predictions made from Internet usage data in order to create novel features and a predictive framework that has generalized value, are performed to predict student's study performance. Thirteen scenarios were created as part of the set. Both student assignment grades and the VLE activity log, which documented the students' interactions with the VLE system, were included in the dataset used for this review. The LISp-Miner tool was used for implementation. They came to the conclusion that both approaches could yield insightful information about the dataset. A graphical model based on Markov Chains can assist in visualizing the information, making it easier to comprehend. The intervention plan receives sub-station support from the patterns that were extracted using the previously described techniques. Predicting student performance throughout their academic career is made easier by analyzing behavioral data.
[0005] FIG. 2 illustrates a method of the workflow of the designed model according to an embodiment herein. In some embodiments, the information about faculty, departments, and students' midterm and final exam scores were identified as features in the dataset. Every measure includes student-related data. Under the "dataset" heading, the grade variables for the midterm and final exams were described. The department variable represents departments within faculties, while the faculty variable represents faculties at Kirsehir Ahi Evran University. The midterm, faculty, and departmental data were identified as independent variables in the model's development, while the final was identified as the dependent variable. With the exception of mobile connection frequency and total connection frequency, the majority of features also demonstrate notable differences between students with high and low scores. In terms of download volume, upload volume, and overall volume, failed students perform noticeably worse than passed students. However, passed students are reported to have a much higher connection frequency. Although it is not statistically significant, they also spend more time on the Internet on average each day than their counterparts. The dataset consists of 1347 instances of writing assignment marks, each of which has four features and four attributes for written assignment scores. Naive Bayes (NB), Neural Network (NN), and WINDOW are the three algorithms that were utilized to construct the system using a combinational incremental ensemble. The way the system operates is that the models are first trained using the training set, and then they are tested using the test set. All three classifiers make predictions when a new observation instance is received.
[0006] In some embodiments, the confusion matrix displays the dataset's current state as well as the proportion of accurate and inaccurate model predictions. The matrix of confusion. The number of instances that are correctly and incorrectly classified indicates the model's performance. The columns display the model's estimation, while the rows display the actual numbers of the test set's samples. Given the strong correlation, it is possible to use Internet usage patterns as feature classes to forecast students' academic success. As supervised learning methods, we used DT, SVM, and NN. Due to the robustness of Decision Tree-based estimators to arbitrary scaling of the data, the detailed usage features were normalized to promote the computing efficiencies of SVM and NN. They employed Weka software as a tool for implementation. Error rate, learning time, and accuracy are used to assess the classifiers. With a training time of less than one second and high error rates, the NB achieves a high accuracy score of 76.65%. Additionally, data mining techniques for predicting student performance are reviewed by Baradwaj and Pal. In order to extract useful rules from the dataset, they look into the accuracy of DT. Purvarichal University in India provided the dataset they used for their review, which included the records of 50 students, each of whom had eight attributes.
[0007] Analytical learning, student behavior prediction, and innovative approaches to educational policy are just a few of the improvements that come from processing educational data. In addition to enabling education authorities to create data-driven policies, this extensive data collection will serve as the foundation for artificial intelligence-based learning process software. Notably, the way that students use the Internet during the various time slots throughout the day offers valuable insights into their habits and behaviors. Our previous study carefully examined how people use the Internet throughout the day and found that cutting off access at 12 a.m. would improve students' academic performance. Additionally, a recent study demonstrated that early morning internet browsing for entertainment was linked to poor academic performance in people with Internet addiction. Students are enrolled in online courses as part of e-learning, a rapidly expanding and sophisticated educational approach. EDM is fully utilized by e-learning platforms like Massive Open Online Courses (MOOC), Learning Management Systems (LMS), and Intelligent Tutoring Systems (ITS) in the development and construction of recommender, adaptive, and automatic grading systems.
[0008] In some embodiments, a classification problem's performance is assessed using the AUC-ROC curve. AUC-ROC describes how well a model predicts and is a commonly used metric to assess the performance of machine learning algorithms, particularly when there are unbalanced datasets. The predictive models were fed datasets with various feature groups. Then, in order to determine the predictive outcomes of passed and failed as well as high-score and non-high-score independently, we estimated the held-out data. With the same number of students in each of those two classes, the baseline accuracy is 50%. Three feature selection techniques-genetic search, greedy step-wise, and best first method-were used to extract 53 features in total from the dataset. The features with low information gain were then removed, reducing the total number of features. Classification, feature selection, and algorithm implementation were done using the Weka data mining tool. The findings indicated that when the classifiers were assessed using the 10-k fold cross-validation and percent split methods, they achieved an accuracy of 88% to 93%.
, Claims:CLAIMS

I/We Claim:
1. A method for deep learning-based prediction of academic performance of students in higher education influenced by internet usage, wherein the method comprises;
gathering extensive data on students' internet usage patterns time on specific websites, online study habits, etc. and academic performance grades, attendance, assignment completion;
cleaning and preprocess data, standardizing timestamps, and encoding categorical data as needed. aggregate usage metrics based on time intervals to ensure a consistent input format;
developing relevant features from internet usage data e.g., total study time, time on social media, academic-related websites to enhance the deep learning model's accuracy;
selecting and design appropriate deep learning architectures that can handle sequential data effectively;
training the model using a labeled dataset e.g., known academic outcomes, optimizing hyperparameters for the best predictive performance;
testing the model with unseen data, calculating metrics like accuracy, precision, and recall to evaluate its effectiveness in predicting academic performance based on internet usage;
deploying the model to predict academic performance, and use interpretation techniques to understand the impact of different internet usage patterns on model predictions; and
using model outputs to provide personalized recommendations or interventions to students, helping them balance internet usage with academic goals.

Documents