Comparative analysis of machine learning algorithms for personalising educational content in distance learning

Abstract

The aim of this research was to conduct a comprehensive evaluation of the effectiveness of machine learning algorithms for the task of personalised educational content recommendation in distance education systems. The study was of a theoretical-experimental nature and was performed using a synthetic dataset comprising 10,000 student profiles, constructed based on the structural characteristics of leading distance learning platforms. The dataset covered three groups of features: demographic, behavioural, and content-related, replicating key patterns of student interaction with the learning environment. A comparative analysis of the effectiveness of the Support Vector Machine (SVM), Decision Tree, Random Forest, and Multilayer Neural Network methods revealed clear quantitative differences between the models. The highest classification results were obtained for the Neural Network (accuracy = 0.91; F1-score = 0.90). The ensemble-based Random Forest model provided high stability and accuracy (accuracy = 0.89; F1-score = 0.87). The Support Vector Machine method showed balanced performance (accuracy = 0.86; F1-score = 0.83), while the Decision Tree exhibited the lowest effectiveness (accuracy = 0.72; F1-score = 0.70), confirming the limitations of interpretable models in multidimensional data. An additional systematic analysis, performed using semi-quantitative indices for six algorithm characteristics, reflected the overall suitability of the models for personalisation: the Neural Network scored 23 points, Random Forest – 21 points, SVM – 19 points, Decision Tree – 17 points. These scores align with the classification metrics and confirm the advantages of models with pronounced non-linearity and ensemble structure. The Multilayer Neural Network demonstrates the highest efficacy for deep content personalisation, Random Forest serves as a universal model for large-scale educational platforms, the Support Vector Machine method is optimal for courses with clearly segmented student groups, while the Decision Tree is advisable to use as an interpretable analytical module. The practical significance of the study lies in forming a scientifically grounded approach to ing algorithms for building adaptive educational trajectories and improving the effectiveness of digital education.