Leveraging Artificial Intelligence to Recommend Personalized Learning Pathways: A Case Study on Secondary Education Students

Abstract

This study aims to develop an explainable artificial intelligence (XAI)–based educational recommendation system tailored for secondary school students. The system integrates classification and regression models to suggest personalized learning paths while estimating the confidence level of each recommendation. The dataset used was obtained from EdNet, comprising data for over 50,000 students including academic performance, educational level, interests, and learning pathways. The study followed an applied quantitative methodology involving data preprocessing, feature encoding, TF-IDF for text processing, model building, and the development of a user interface using Gradio. Four classification algorithms were evaluated: Decision Tree, Random Forest, KNN, and XGBoost. XGBoost outperformed others with an accuracy of 89%. For regression, LightGBM achieved the highest performance (R² = 0.9997, MAE = 0.000008). SHAP was utilized to interpret model outputs and provide visual explanations of the recommendation logic. The results demonstrated that incorporating explainability into educational recommender systems significantly increases users’ trust and engagement. The study recommends enriching the input data with behavioral indicators, testing the model in diverse learning contexts, and deploying the system as an interactive platform aligned with Saudi Vision 2030’s digital education goals.