Development of an Offline-Capable Machine Learning Dashboard for Early Academic Risk Prediction in Resource-Constrained Schools

Abstract

One of the challenges particularly facing schools in resource-constrained settings is early identification of students struggling academically. This challenge is mainly caused by limited educational resources and unreliable internet connectivity making these schools unable to provide the necessary support at the right time. To address this problem, this study developed an offline- capable machine learning dashboard for early academic risk prediction in public secondary schools in Onicha Local Government Area, Ebonyi State, Nigeria. Data containing variables such as; demographic, socio-economic, behavioural, and motivational variables with academic performance categorized as high or low, were collected from 378 public secondary school students. A Random Forest model was trained on the dataset and model achieved an accuracy of 89.5%, precision of 89.1%, recall of 85.4%, an F1-score of 87.2%, and an AUC of 0.949, demonstrating high predictive performance. The model after training was deployed as an interactive dashboard using Streamlit, a Python framework that can be run completely offline on basic school computers without Internet connectivity or special technical expertise. Teachers can input student data into the dashboard and get immediate predictions of academic risk along with actionable recommendations for intervention. Key features include: real time prediction using the trained model, visualization of top influential predictive factors, automated generation of affordable intervention strategies, offline deployment, and batch prediction through uploading CSV files. The system we propose demonstrates that accurate, interpretable, and deployable machine learning solutions can effectively support educational decision-making in limited resource environments, providing a practical framework for accessible educational data mining worldwide.

Keywords

References