AN EXPLAINABLE MULTI-AGENT LEARNING ANALYTICS PLATFORM FOR STUDENT PERFORMANCE PREDICTION, RISK ASSESSMENT, AND PERSONALIZED EDUCATIONAL SUPPORT

Abstract

In recent years, Artificial Intelligence (AI) has been widely adopted in educational analytics to support student performance prediction and data-driven decision-making. Nevertheless, concerns regarding transparency, explain ability, and personalized support continue to restrict the practical adoption of many existing educational AI systems. This study proposes an Explainable Multi-Agent Learning Analytics Platform that combines machine learning, explainable AI (XAI), and intelligent agent technologies to overcome these challenges. The proposed framework comprises 6 specific agents: Prediction, Risk Assessment, Recommendation Generation, Knowledge Graph, Retrieval-Augmented Generation (RAG), and Digital Twin Simulation. The global SHAP and local LIME are used to give global and local explanations to the prediction results, which can be very useful for understanding the workings of the system. The xAPI-Edu-Data (Kalboard 360) dataset has been evaluated using four machine learning models: Random Forest, XGBoost, Light GBM, and Cat Boost. The baseline prediction model achieved 87.0% accuracy, whereas the full multi-agent framework increased overall accuracy to 89.9% and F1 score to 89.9%. Attendance behavior and risk factors were the most important factors that were identified by feature importance analysis. A high-risk group of 26% of students, a medium-risk group of 39% of students, and a low-risk group of 35% of students were identified through a risk assessment, which informs instructional and learning strategies that can be implemented as early as possible in the school year. Additionally, SHAP showed better explanation stability (0.85) than LIME (0.72), which means the explanations are more stable in the context of educational decisions. The proposed framework is an effective and interpretable solution for personalized learning analytics and the development of reliable AI-based educational systems.