VIETNAM NATIONAL UNIVERSITY, HANOI INTERNATIONAL SCHOOL GRADUATION PROJECT PREDICTING STUDENTS' PERFORMANCE OF PRE-ENGLISH COURSE BY USING NEURAL NETWORK – A CASE STUDY IN INTERNATIONAL SCHOOL – VIETNAM NATIONAL UNIVERSITY, HANOI Student’s name: Le Quynh Hoa Hanoi – 2023 VIETNAM NATIONAL UNIVERSITY, HANOI INTERNATIONAL SCHOOL GRADUATION PROJECT PREDICTING STUDENTS' PERFORMANCE OF PRE-ENGLISH COURSE BY USING NEURAL NETWORK – A CASE STUDY IN INTERNATIONAL SCHOOL – VIETNAM NATIONAL UNIVERSITY, HANOI SUPERVISOR: Dr. Nguyen Quang Thuan STUDENT: Le Quynh Hoa STUDENT ID: 20070930 FACULTY: Faculty of Applied Sciences MAJOR: Business Data Analytics CLASS: BDA2020A Hanoi – 2023 ACKNOWLEDGEMENTS I would like to express my deepest gratitude to my thesis advisor, Dr. Nguyen Quang Thuan, for his invaluable guidance, support, and expertise throughout the research process. His insightful feedback, dedication, and encouragement were instrumental in shaping this thesis.
I would like to thank the Applied Sciences faculty for providing a conducive environment for learning and research throughout my 3 years at the university. The resources, facilities, and opportunities provided by the institution were crucial in conducting this study. And the Academic department for giving us access to the students’ databases data used in this thesis. I would also like to thank my friends and colleagues for their support, discussions, and willingness to provide assistance whenever needed.
Their camaraderie and shared experiences made this journey memorable. Finally, I extend my gratitude to all the participants who generously contributed their time, insights, and data to this research. With their involvement, this study was possible. This thesis is a culmination of many individuals' collective efforts and support, and I am humbled by their contributions.
Thank you all for being integral to my academic pursuit and helping me achieve this milestone. Hanoi, 2023 Le Quynh Hoa 2 DECLARATION I, Le Quynh Hoa, hereby declare that this thesis, titled Predicting Students' Performance of pre-English Course by Using Neural Network – A case study in International School – Vietnam National University, Hanoi, submitted partially to fulfill the requirements for the degree of Business Data Analytics at International School, Vietnam National Univerisity, Hanoi, is my original work. This thesis's ideas, data, and information are sourced and referenced appropriately. I further declare that: (1) Any assistance received for this thesis, including supervisor guidance, discussions with peers, and use of external sources, is duly acknowledged and properly cited.
(2) Any previously published work or contributions by others used in this thesis have been appropriately acknowledged with full citations. (3) The thesis has not been submitted, in part or in whole, for any other degree or qualification at any other university or institution. (4) The research conducted for this thesis has been carried out with ethical considerations in mind, ensuring the protection of participants' rights, privacy, and confidentiality. The necessary ethical approvals and permissions were obtained for the collection and analysis of data.
(5) Any conflicts of interest related to this thesis have been disclosed and appropriately managed. This thesis represents my independent thinking and understanding of the subject matter without falsifying, fabricating, or manipulating data or results. All images, figures, tables, and other materials used in this thesis are either created by myself or sourced from the public domain or with proper permissions and acknowledgments from the original creators. I understand that any violation of the above declarations may result in consequences as per the academic policies and integrity guidelines of International School, Vietnam National Univerisity, Hanoi.
Hanoi, 2023 Le Quynh Hoa 3 TABLE OF CONTENTS TABLE OF CONTENTS. 4 LIST OF ABBREVIATIONS. 6 LIST OF FIGURES. 6 LIST OF TABLES.
Background and Significance. Previous Approaches in Student Performance Assessment. Data-Driven Approaches. Supervised Learning in Student Performance Assessment.
Application of Deep Learning in Student Performance Assessment. Advantage of Using Neural Network Approach for Multiclass Classification in Predicting Student Performance. 14 CHAPTER 2: FUNDAMENTALS OF ARTIFICIAL NEURAL NETWORKS. Fundamentals of artificial neural networks.
Feedforward Neural Network Architecture. Neural networks’ parameters. Propagation in Neural Network. Backpropagation & Optimization Algorithm in Neural Network.
Techniques and Methodology. Cross-Validation Techniques. Models and Parameters:. Evaluation and Performance Metrics.
26 CHAPTER 3: APPLYING NEURAL NETWORKS FOR PREDICTING TIME TO ACHIEVE B2 CERTIFICATION. Preprocessing Steps Applied to the Dataset. Data selection & Labeling. Handling missing data.
Checking the correlation. Perform Encoding and Data Scaling for Training Variables. Results and Analysis. 51 5 LIST OF ABBREVIATIONS Adam.
Adaptive Moment Estimation ANN. Aritificial Neural Network AUC. Area Under the Curve CNN. Convolutional Neural Network EDM.
Education Data Mining FFNN. Feedforward Neural Network NN. Neural Network OHE. One-hot encoding PE.
Pre-English ReLU. Rectified Linear Unit RMSprop. Root Mean Square Propagation ROC. Receiver Operating Characteristic SMOTE.
Synthetic Minority Oversampling Technique VNUIS. International School, Vietnam National University LIST OF EQUATIONS Equation 1. Function of employed metrics in classification. 26 LIST OF FIGURES Figure 1.
General Structure of Artificial Neural Network with Two Hidden Layers15 Figure 2. Data Processing in a Neuron (Source: Medium). Weight and Bias in a neuron (Source: InfoWorld). Common used activation functions (Source: AI Wiki).
Visualization of ReLU Function. The forward propagation of a neural network (Source: ResearchGate). The role of loss function after a observation (Source: Medium). The Backpropagation process (Source: Analytics Arora).
The case of choosing learning rate in NN (Source: Jeremy Jordan). Epoch and Batch size (Source: Chen-Jun Wang). Example of Train-Test Loss Learning Curve (Source: Machine Learning Mastery). Visualization of MultiClass ROC, rechieve from Keras Library.
Extraction of dataset. Visualization of missing rows in dataset. The frequency of 2nd subject after imputing mode imputation. Correlation between numerical features and target outcome.
Numerical training set after implementing StandardScaler. Number of Startified data after Train - test split. Class Distribution before and after SMOTE. Learning curve of base model with out birthplace.
Learning curve of the best base model. Trying with different epochs. Differences between RMSprop and Adam optimizer. Learning curve of applying RMSProp optimizer.
Examining different hidden layers. The ROC curve illustrates the trade-off between the true positive rate (sensitivity) and the false positive rate (1 - specificity) for each class. 47 LIST OF TABLES Table 1. Number of intake classes after labeling.
Numbers of train - validation - test set after splitting. Examining different datasets into the FFNN model. Confusion matrix showing the classification results for each class. Classification report displaying the precision, recall, F1-score, and support for each class.
46 7 ABSTRACT Nowadays, Education Data Mining (EDM) enables educators and policymakers to gain better insight into students’ needs, leading to informed decisions about teaching methods and curriculum adjustments. As education data mining continues to rise, applying Neural networks is becoming increasingly popular due to its ability to handle complex relationships between variables and identify hidden patterns, ultimately leading to more accurate predictions than other algorithms. The thesis studies Neural network: A case study in International School – Vietnam National University, Hanoi, is applied to illustrate neural networks' application in classifying problems and predicting student performance. In the progress of the training model, preparation steps take high account of the project to help the model obtain the input data; additionally, adjusting the parameters and hyperparameters is another crucial step of the Neural network, making the Neural network stand for others.
Results indicated that the placement test score and birthplace were the most influential predictors for student performance, challenging some previously held assumptions. This study provides valuable insights for improving the effectiveness of post- secondary undergraduate programs worldwide. 8 CHAPTER 1: INTRODUCTION This chapter describes the research topic of the thesis - student performance prediction, emphasizing its significance in education. The chapter depicts traditional and data-driven approaches in student performance assessment, including supervised learning and deep learning techniques.
On the other hand, providing an overview of the application of the algorithm in VNU-IS setting and lists down the question for the study. Background and Significance In the education industry, predicting student performance is necessary when the policymaker analyzes the learning outcomes of a course or a curriculum, oversees instructional strategies, and develops educational policies. Classifying students' accurate performance across multiple classes helps personalized learning, early intervention, and educational decision-making. However, this task faces challenges since the subjective nature of performance assessment, the need for consistent and fair classification systems, and the impact of class imbalance on classification effectiveness.
In education data mining (EDM), researchers have used data analytics to extract valuable insights from large datasets to improve educational outcomes. Recent studies have shown the effectiveness of data mining techniques in forecasting student performance. For example, clickstream interactions in Massive Open Online Courses have been examined to predict student attrition, as demonstrated in the study by Sinha et al. Additionally, automated student models have been employed to predict performance, although the specific reference for this work is not provided in the statement.
These findings have significant implications for educators, policymakers, and students. Early identification of at-risk students enables targeted interventions, improving their chances of success. Policymakers can allocate resources based on performance predictions to support effective programs and schools. Furthermore, students themselves can benefit from knowing their predicted performance, using it as motivation to work harder and make informed decisions about their education.
Categorizing student performance into multiple classes or levels presents various challenges. In this thesis, our focus is on the application of student performance 9 prediction using EDM within a specific context. We will investigate the placement process of the pre-English (PE) course at International School - Vietnam National University and explore the factors that influence it. Context The faculty of Applied Linguistics at International School - Vietnam National University (VNUIS) is looking for ways to improve the accuracy and efficiency of the current process for placing students into the pre-English course.
A potential solution being considered is to utilize EDM method, which uses data-driven techniques to predict students’ academic performance. The placement test evaluates each student's language skills through a writing, reading, and listening test. However, accurately classifying students based on these marks has proven challenging. By using EDM techniques, the department can analyze various factors influencing student performance, such as academic performance, language test scores, demographic information, and subsequent English proficiency levels.
This will help the department establish a more effective and equitable classification system. Enhancing the placement process for pre-English courses is essential to ensure students are assigned to classes matching their language skills, leading to an enhanced learning experience. The study’s outcomes will inform decision-making regarding the necessity of taking the PE course and its impact on students’ English proficiency levels. The application of EDM techniques for student performance prediction holds promise for enhancing the classifying process’ overall effectiveness and improving the student's educational experience.
Previous Approaches in Student Performance Assessment 1. Traditional Approaches Traditional approaches have been widely employed in student performance assessment, including subjective evaluation by teachers, exams, and assignments. These approaches have been used for many years due to their familiarity and ease of implementation (2). However, they suffer from several limitations.
One major drawback is the subjectivity of the evaluation process, as different teachers may have varying interpretations of performance indicators (3). This can lead to inconsistent assessment outcomes and potential biases. 10 Another challenge with traditional approaches is their limited ability to provide timely feedback and personalized learning experiences. Since traditional assessments are often conducted at fixed intervals, such as semester exams, which may not capture the dynamic progress of students.
Additionally, traditional approaches focus on outcomes rather than the learning process itself. Data-Driven Approaches The use of data-driven approaches in student performance assessment has become more prevalent with the rise of educational data mining, machine learning, and predictive analytics.