Foundations for an Enhanced XAI System: A Systematic Literature Review on Optimising LIME for Distance Learning Environments

Explainable Artificial Intelligence (XAI) has become increasingly essential in distance learning systems, where predictive models are used to support assessment, personalised feedback, learner monitoring, and early-warning interventions. Local Interpretable Model-Agnostic Explanations (LIME) is a widely adopted XAI techniques due to its model flexibility and ease of implementation. However, LIME suffers from methodological limitations and limited global interpretability, that undermine its suitability for high-stakes educational applications. This systematic literature review analysed research published on LIME optimisation, categorising existing enhancements into six major strategy groups: deterministic and uncertainty-based sampling, data-aware perturbations, nonlinear surrogate models, global and aggregated explainability methods, visual and relational interpretation enhancements, and hybrid or optimisation-driven frameworks. Findings show that while numerous variants significantly improve stability, fidelity, or interpretability within their respective domains, most remain domain-specific, computationally demanding, and insufficiently evaluated through human-centred or educational lenses. Only a few studies explore LIME optimisation within learning analytics, leaving a substantial gap in applying enhanced XAI methods to distance learning environments. The review identifies unresolved challenges like the need for multimodal and sequential data handling, unified optimisation across multiple objectives, scalable real-time implementation, and empirically validated human interpretability. Insights from literature inform a set of design requirements for an improved LIME-based XAI system designed for distance learning.