Abdelali Zakrani

The analysis of students’ learning traces in digital environments enables a better understanding of the factors influencing their academic performance. This paper proposes a hybrid predictive model combining Machine Learning (ML), Deep Learning (DL), and Generative Artificial Intelligence (GenAI) to predict students’ academic performance (GPA) by leveraging traces from six dimensions: demographic, cognitive, social, emotional, contextual, and normative. First, the Random Forest algorithm is used to select the most relevant features. Then, a combined model based on MLP, LSTM, and XGBoost is trained to optimize prediction accuracy. Generative AI (DeepSeek) is integrated to enrich contextual data and provide personalized recommendations. Model decisions are interpreted using SHAP, allowing for explainability of the predictions. The model is evaluated on a real dataset of 1,002,393 records, built from learning traces collected from student interactions with courses hosted on a Moodle platform. The results highlight a significant improvement in predictive performance (a coefficient of determination of 87%, an average precision of 89.2%, an overall recall of 87.5%, an F1-score of 88.3%, and an accuracy of 90.4%) compared to traditional approaches. This work underscores the value of hybrid techniques for advanced educational data analysis. The results open promising perspectives for integrating intelligent learning support systems, as well as for adapting and extending the model to other educational contexts, learning platforms, and academic levels.

Faced with the rise of online learning platforms, predicting learners’ academic performance has become a major concern to personalize and enhance educational journeys. However, traditional predictive models struggle to effectively integrate emotional and social factors. This article introduces a hybrid predictive model that combines random forests (RF) for selecting the most relevant features and multiple regression (MR) to forecast academic performance. The data is sourced from three online learning platforms and encompasses both implicit traces (learner interactions and behaviors) and explicit traces (demographic characteristics). Following a selection and merging process, the final dataset comprises 1,003,392 records and 42 features, categorized into six types of indicators: cognitive, emotional, social, normative, contextual, and demographic. The results demonstrate that this hybrid model outperforms traditional approaches and other machine learning (ML) techniques in terms of predictive accuracy, achieving an R² of 0.9372 and a root mean square error (RMSE) of 0.1022. The incorporation of explicit and implicit traces helps better capture the intricate interactions among the different data dimensions, significantly enhancing prediction quality. This work represents a notable advancement in the field of academic performance prediction. It also sheds light on challenges associated with the increasing complexity of models, paving the way for future research to develop more generalizable approaches.

The emergence of online learning has sparked increased interest in predicting learners’ academic performance to enhance teaching effectiveness and personalized learning. In this context, we propose a complex model APPMLT-CBT which aimes to predict learners’ performance in online learning settings. This systemic model integrates cognitive, social, emotional, contextual, and normative aspects to predict the learners’ performance in online learning environment. This model, based on Competency-Based Learning Traces, takes a holistic approach by integrating various data reflecting knowledge acquisition and skills development. By Taking into account the exchanges among the learners, as well as the interactions with their teachers and the complexity of their online learning environment, the model aims to provide accurate and informed predictions of academic performance. This study provides a detailed overview of the APPMLT-CBT model, its data collection methodology, and discusses its potential implications for online teaching. Results suggest that the model can serve as a robust framework for improving online teaching and learning while offering a deep understanding of the underlying mechanisms of online learning.

With the rapid evolution of online learning environments, the ability to predict students' academic performance has become crucial for personalizing and enhancing the educational experience. In this article, we present a predictive model based on machine learning techniques, designed to be integrated into online learning platforms using the competency-based approach. This model leverages features from four key dimensions: demographic, social, emotional, and cognitive, to accurately predict learners' academic performance. We detail the methodology for collecting and processing learning traces, distinguishing between explicit traces, such as demographic data, and implicit traces, which capture learners' interactions and behaviors during their learning process. The analysis of these data not only improves the accuracy of performance predictions but also provides valuable insights into skill acquisition and learners' personal development. The results of this study demonstrate the potential of this model to transform online education by making it more adaptive and focused on individual learners' needs.

Breast cancer is a significant contributor to female mortality, emphasizing the importance of early detection. Predicting breast cancer accurately remains a complex challenge within medical data analysis. Machine learning (ML) algorithms offer valuable assistance in decision-making and diagnosis using medical data. Numerous research studies highlight the effectiveness of ML techniques in improving breast cancer prediction. Feature selection plays a pivotal role in data preprocessing, eliminating irrelevant and redundant features to minimize feature count and improve classification accuracy. This study focuses on optimizing breast cancer diagnostics through feature selection methods, specifically genetic algorithms (GA) and particle swarm optimization (PSO). The research involves a comparative analysis of these methods and the application of a diverse set of ML classification techniques, including logistic regression (LR), support vector machine (SVM), decision tree (DT), and ensemble methods like random forest (RF), AdaBoost, and gradient boosting (GB), using a breast cancer dataset. The models' performance is subsequently evaluated using various performance metrics. The experimental findings illustrate that PSO achieved the highest average accuracy, reaching 99.6% when applied to AdaBoost, while GA attained an accuracy rate of 99.5% when employed with both AdaBoost and RF.

Computing advances in data storage are leading to rapid growth in large-scale datasets. Using all features increases temporal/spatial complexity and negatively influences performance. Feature selection is a fundamental stage in data preprocessing, removing redundant and irrelevant features to minimize the number of features and enhance the performance of classification accuracy. Numerous optimization algorithms were employed to handle feature selection (FS) problems, and they outperform conventional FS techniques. However, there is no metaheuristic FS method that outperforms other optimization algorithms in many datasets. This motivated our study to incorporate the advantages of various optimization techniques to obtain a powerful technique that outperforms other methods in many datasets from different domains. In this article, a novel combined method GASI is developed using swarm intelligence (SI) based feature selection techniques and genetic algorithms (GA) that uses a multi-objective fitness function to seek the optimal subset of features. To assess the performance of the proposed approach, seven datasets have been collected from the UCI repository and exploited to test the newly established feature selection technique. The experimental results demonstrate that the suggested method GASI outperforms many powerful SI-based feature selection techniques studied. GASI obtains a better average fitness value and improves classification performance.