Abstract

Preventing student dropouts in small rural schools remains a major challenge due to the opaque (“black box”) nature of traditional artificial intelligence models and the difficulties of adapting them to resource-constrained environments.

This study proposes the AI-based Early Warning and Dropout Surveillance (AI-EWDS) framework, a novel intelligent architecture engineered to improve predictive transparency and contextual adaptability.

The primary technical contribution of this study is the integration of an automated AI agent architecture with Explainable Artificial Intelligence (XAI), utilizing SHAP and feature importance techniques to demystify the decision-making process and make AI reasoning fully interpretable to teachers.

Additionally, the framework employs a Human-in-the-Loop (HITL) mechanism, establishing a continuous feedback loop to refine the model’s learning process and align it with specific local school contexts.

Evaluation results indicate high system suitability, with an average overall score of 4.73 out of 5.00 in the Learning Quality and Adaptive Intelligence dimensions and 4.74 specifically in the Learning Quality dimension.

This work is licensed under a Creative Commons Attribution 4.0 License.