Knowledge Distillation in RNN-Attention Models for Early Prediction of Student Performance

Proceedings of the 37th ACM/SIGAPP Symposium on Applied Computing, Journal Year: 2025, Volume and Issue: unknown, P. 64 - 73

Published: March 31, 2025

Educational data mining (EDM) is a part of applied computing that focuses on automatically analyzing from learning contexts. Early prediction for identifying at-risk students crucial and widely researched topic in EDM research. It enables instructors to support stay track, preventing student dropout or failure. Previous studies have predicted students' performance identify by using machine collected e-learning platforms. However, most aimed utilizing the entire course after finished. This does not correspond real-world scenario may drop out before ends. To address this problem, we introduce an RNN-Attention-KD (knowledge distillation) framework predict early throughout course. leverages strengths Recurrent Neural Networks (RNNs) handling time-sequence at each time step employs attention mechanism focus relevant steps improved predictive accuracy. At same time, KD compress facilitate prediction. In empirical evaluation, outperforms traditional neural network models terms recall F1-measure. For example, it obtained F1-measure 0.49 0.51 Weeks 1--3 0.61 1--6 across all datasets four years university Then, ablation study investigated contributions different knowledge transfer methods (distillation objectives). We found hint loss hidden layer RNN context vector module could enhance model's students. These results are researchers employing deep models.

Language: Английский

---

Harnessing variable reduction approach with deep recurrent neural network for student’s academic performance analysis

Alexandria Engineering Journal, Journal Year: 2025, Volume and Issue: 118, P. 393 - 405

Published: Jan. 23, 2025

Language: Английский

---

PREDICTING ACEDEMIC PERFORMANCE OF HIGHER EDUCATION STUDENTS BASED ON THEIR POTENTIAL INTENTION AND BEHAVIOUR ANALYSIS USING AI

SSRN Electronic Journal, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 1, 2025

Language: Английский

---

Beyond the Classroom: Understanding the Evolution of Educational Data Mining With Key Route Main Path Analysis

Computer Applications in Engineering Education, Journal Year: 2025, Volume and Issue: 33(2)

Published: Feb. 18, 2025

ABSTRACT Educational data mining (EDM) enhances the educational system by uncovering hidden patterns of academic data. The discipline EDM has grown rapidly and produced numerous publications, leading to knowledge dissemination among researchers. This research aims understand field literature examining citation network significant publications. utilizes a quantitative approach based on main path analysis (MPA) analyze 1009 Web Science (WoS) publications between 1988 2023. study uncovers 22 that have shaped diffusion trajectories EDM. reveals undergone three phases evolution, each which represents substantial shift in focus: automated adaptation, leveraging human decision, advanced predictive analytics. Unlike previous reviews, this applies novel using multiple global MPA, five key sub‐research areas: student performance, early warning, learning behavior, transfer learning, dropout. Notably, recent trends emphasize growing focus performance. primary contribution paper lies its comprehensive mapping EDM's developmental trajectory, offering an understanding diverse trends. By elucidating these emerging areas, not only enriches existing but also identifies unexplored topics can guide future directions, distinguishing itself from other reviews more systematic data‐driven field's evolution.

Language: Английский

---

Artificial Intelligence in Computer Programming Education: A Systematic Literature Review

Computers and Education Artificial Intelligence, Journal Year: 2025, Volume and Issue: unknown, P. 100403 - 100403

Published: April 1, 2025

Language: Английский

---

Mapping the Landscape of Generative Artificial Intelligence in Learning Analytics: A Systematic Literature Review

Published: Jan. 1, 2025

Language: Английский

---

Use of Deep Learning Techniques for Classification and Prediction of High School Students Needing Academic Support

Lecture notes in networks and systems, Journal Year: 2025, Volume and Issue: unknown, P. 225 - 234

Published: Jan. 1, 2025

Language: Английский

---

Feature selection techniques and classification algorithms for student performance classification: a review

International Journal of Power Electronics and Drive Systems/International Journal of Electrical and Computer Engineering, Journal Year: 2024, Volume and Issue: 14(3), P. 3230 - 3230

Published: April 4, 2024

The process of categorizing students’ performance based on input data, encompassing demographic information and final exam results, is recognized as student classification. Educational data mining has gained traction in assessing performance. However, this study entails the need to analyze diverse attributes within an educational institution by using techniques. This thoroughly examines both previous current methodologies presented researchers, addressing two main aspects: preprocessing classification algorithms applied Data specifically delves into exploration feature selection techniques, three types search methods. These techniques aim identify most significant features, eliminate unnecessary ones, reduce dimensionality. In addition, play a crucial role or predicting Models such k-nearest neighbors (KNN), decision tree (DT), artificial neural networks (ANN), linear models (LR) were scrutinized their prior research. Ultimately, highlights potential for further like gain, Chi-square, sequential selection, particularly when new datasets online learning activities, utilizing variety algorithms.

Language: Английский

---

Student performance classification: a comparison of feature selection methods based on online learning activities

International Journal of Power Electronics and Drive Systems/International Journal of Electrical and Computer Engineering, Journal Year: 2024, Volume and Issue: 14(4), P. 4675 - 4675

Published: June 4, 2024

The classification of student performance involves categorizing students' using input data such as demographic information and examination results. However, our study introduces a novel approach by emphasizing online learning activities rich source. To avoid misinterpretation during the classification, we therefore presented comparing several feature selection (FS) methods combined with artificial neural network (ANN), for classifying students’ based on their activities. At first, focused tackling issue missing values implementing cleaning variance threshold. Feature techniques were implemented which encompass both filter-based (information gain, chi-square, Pearson correlation) wrapper-based, sequential (forward backward) techniques. In stage, multi-layer perceptron (MLP) was used default hyperparameters 5-fold cross-validation along synthetic minority oversampling technique (SMOTE) also applied to each method. We evaluated method's key metrics: accuracy, precision, recall, F1-score. outcomes highlighted gain top-performing methods, all achieving 100% accuracy. This research underscores potential leveraging robust within specified constraints.

Language: Английский

---

A Multi-View Deep Learning Approach for Enhanced Student Academic Performance Prediction

Deleted Journal, Journal Year: 2024, Volume and Issue: 31(6s), P. 293 - 304

Published: Aug. 15, 2024

Educational institutions are utilizing Deep Learning (DL) techniques to develop predictive systems that identify students at risk of underperforming based on historical academic data patterns, thereby enhancing their educational outcomes through targeted interventions. From this outlook, an Ensemble Generative Adversarial Network with a Student Accomplishment prediction using the Distinctive DL (EGAN-SADDL) model was designed generate large-scale student and predict achievements. However, integrating heterogeneous kinds into SADDL is complex task that, if not executed properly, may result in failing capture crucial relationships, leading lower performance. Hence, paper proposes EGAN Improved (EGAN-ISADDL) multi-view learning for predicting The main aim learn features from multiple sources, including records, demographic information, social media activity, approach. First, attributes collected, along physiological extracted information posted by students. Second, Long Short-Term Memory Convolutional Neural (LSTM-DCNN) Recursive (ReNN) models receive these parallel, extracting intermediate views. Third, classifier jointly learns each set students' performance, enabling early identification at-risk high accuracy. Finally, experiments conducted dataset 50,000 records demonstrate EGAN-ISADDL attains 96.28% accuracy compared existing single-view models.

Language: Английский

---