XPLearn: An Explainable AI Framework for Personalized Learning Path Recommendation

Abstract

ABSTRACT

Personalized learning platforms are generating vast amounts of learner data, yet many recommendation systems still lean on rigid, rule-based logic that overlooks curriculum progression, subject continuity, and how individual learners actually behave. XPLearn was developed to close this gap: an explainable AI framework that pairs curriculum-aware feature engineering with an Enhanced Random Forest (ERF) classifier to recommend personalized learning paths. The framework was tested on a publicly available educational dataset comprising 1,000 learner records, seven predictor variables, and 12 recommendation classes. Five baseline models — Logistic Regression, Support Vector Machine, Random Forest, Gradient Boosting, and a Multi-Layer Perceptron — were benchmarked against the proposed ERF configuration, which draws on 44 engineered features spanning behavioural, structural, completion-history, and encoded learner attributes. Model transparency was assessed through Gini importance and permutation importance. Across the experiments, XPLearn reached 98.4% test accuracy, 98.1% precision, 97.8% recall, and 97.9% F1-score, with a ROC-AUC of 0.981 and an inference time of 6.1 ms. Permutation analysis singled out Level_Progression and Performance_Score as the two most influential predictors. Together, these results point to the value of pairing curriculum-aware representation with interpretable ensemble learning for educational recommendation. That said, before the framework is used outside this study, target-leakage control for the target-adjacent engineered features needs to be explicitly verified, and the experiments rerun with leakage-free features if the audit raises concerns.

Keywords: Explainable Artificial Intelligence; Personalized Learning; Learning Path Recommendation; Educational Data Mining; Random Forest; Permutation Importance