Educational data mining (EDM) is a specialized field within data mining that focuses on extracting valuable insights from academic data across high school and university levels. A common practice in EDM involves predicting students’ grades to identify at-risk individuals and improve the efficiency of academic tasks. This knowledge benefits students, parents, and institutions equally. Early detection enables interventions that improve student performance. The literature presents various prediction strategies, each with its own unique advantages and disadvantages. This study aims to comprehensively evaluate the methods, tools, and applications of machine learning (ML) and data mining (DM) in education. The main goal is to improve the accuracy of predicting academic achievements by employing eight widely recognized ML algorithms: naïve bayes (NB), k-nearest neighbors (KNN), support vector machine (SVM), random forest (RF), logistic regression (LR), extreme gradient boost (XGBOOST), and ensemble voting classifier (EVC). The focus is on improving data quality by eliminating instances of noise. Performance evaluation involves assessing parameters such as accuracy, precision, F-measure, and recall. Incorporating cross-validation and hyperparameter tuning improves classification accuracy. The ML models outperform other ensemble approaches, providing a valuable tool for predicting student performance and assisting educators in making proactive decisions through timely alerts.
19 March 2024 in International Journal of Interactive Mobile Technologies (iJIM)
Despite the explosion of interest in adaptive learning and learning analytics (LA) for higher education (HE), there has been relatively little research integrating educational approaches’ indicators to build an adaptive learning analytics model. Adaptive learning analytics (ALA) models have grown in favor of HE due to their claims of enhancing student learning outcomes, providing personalized learning paths, and allowing students to interact with course material at their own pace. With focus on using data to personalize the learning experience and the environment in which the experience of learning occurs, LA centers on enhancing education through meticulous data analysis, while big data (BD) in education addresses the overarching challenges and opportunities arising from extensive and varied datasets. These concepts are interconnected, where LA and ALA specifically apply BD principles within the educational context. In this paper, we explain some BD concepts used in HE, define the essential perceptions related to LA, and analyze educational approaches to define the fundamental implications. Besides, we try to connect some LA model with educational approaches based on the big educational data collected in order to establish an efficient educational model. We include all steps cited before we try to build an ALA model in HE that resolves the limitations of the oldest models, thus improving the learner’s learning process by adding and treating additional indicators.
24 March 2020 in Journal of Cyber Security and Mobility
Smart grid is a system in which it is possible to use voting-based techniques to resist sabotage of several cyber-attacks. The adaptation of these techniques can be difficult and useless in the case when the malicious resources (i.e., smart meters) of this system can return wrong data in same time; however, the collusion problem is triggered. To detect and resolve the collusive issue, spot-checking technique has been proposed by sending randomly certain number of spotter queries to chosen resources with known correct data in order to estimate resource credibility based on the returned data. This work proposes an original method that resist against collusion attacks by using probability to solving a new spot-checking optimization problem for smart grid systems, with the objective to minimize probability of accepting wrong data (PAWD) while respecting an expected overhead constraint. The proposed solution contains an optimal combination of several parameters, the number of spotter queries sent, the number of resources tested by each spotter query, and the number of resources assigned to run the genuine query. The optimization procedure includes a new method for evaluating performance metrics of PAWD and expected overhead in terms of the total number of query assignments. To demonstrate the proposed optimization problem and solution procedure, we have provided several illustrative examples.
The ease of usage and the convenience of cloud computing come with considerable responsibility. The latter, consists of carefully addressing different security aspects of this technology. The integrity and availability of the outsourced data constitute essential considerations for adopters’ final decisions. However, the most critical factor is the efficiency of integrity checks, which must prioritize restricted-resource data owners without affecting the performance of the Cloud Service Provider. This paper proposes a secure scheme, called Proof of Exponentiation of Dynamic Data Possession ${ \mathsf {PoEDDP} } $ based on ${ \mathsf {RSA} } $ -Accumulators. The proof of concept demonstrates that this scheme is 20 times faster compared to other ${ \mathsf {RSA} } $ -based cryptographic accumulator schemes. It could be improved to achieve great results with proper optimizations on the larger integer multiplication side.
Early-warning systems aim to identify students at risk of failing or withdrawing in time to act. On two modules of the Open University Learning Analytics Dataset, we evaluated a hybrid model in which a Transformer encoder summarises ten weeks of virtual learning environment activity and XGBoost combines this representation with demographic and early-assessment features. Preprocessing, encoder training and tuning were nested inside student-level cross-validation with equal tuning budgets, and results are reported primarily for the 12,436 registrations not withdrawn by day 70. On this cohort, the hybrid reached an area under the ROC curve (AUC) of 0.825. It outperformed static-only and representation-only models but not gradient boosting on static features with aggregated activity (AUC 0.822) or raw weekly counts (0.832); neither difference was statistically significant. With the representation fixed, a boosted head outperformed logistic and neural heads only when static features were included. Including students who had already withdrawn raised the AUC of every learned configuration by 0.062–0.087, which is more than the differences between competitive models. Predictions were well calibrated, early-assessment variables had the largest individual attributions, and error rates differed across socio-economic groups. The main contribution is a reproducible, leakage-controlled protocol for measuring what a sequence encoder adds.
1 January 2026 in International Journal of Information and Education Technology
The choice of a paramedical specialization suited to students’ academic profiles and preferences is a challenge for Moroccan students. To respond to this challenge, this study evaluated seven tree-based algorithms on real data from 889 nursing students from two Moroccan training institutes to determine the best-performing model to predict the students’ paramedical specialization. We implemented Bayesian hyperparameter optimization with Optuna, stratified 5-fold cross-validation, and final evaluation on an independent test set to ensure a robust and reliable evaluation of the models. The resampling techniques (SMOTE, SMOTEENN, ADASYN, SMOTETomek) were applied only in the training folds to avoid data leakage. Ensemble models (Random Forest, Extra Trees, XGBoost, Gradient Boosting, HistGradient Boosting) surpassed the baseline model (Decision Tree), while AdaBoost’s performance remained weak and unstable in cross-validation and on the test set (test macro F1-score as low as 0.1384), confirming its strong sensitivity to resampling-induced noise. On the test set, Random Forest with SMOTEENN achieved the best overall performance (macro F1-score = 0.8836, macro Receiver Operating Characteristic curve and Area Under the Curve (ROC-AUC) = 0.9879, macro Precision-Recall Area Under the Curve (PR-AUC) = 0.9294), whereas XGBoost with SMOTE delivered near-identical predictive performance (macro F1-score = 0.8828) with a faster inference time (0.0060 s), offering the best deployment trade-off. By integrating these two optimal approaches, an orientation system can accurately predict the specialization best suited to a student’s profile and preferences.