A Machine Learning Framework for Attendance Forecasting and Anomaly Detection in Educational Access Systems
Year: 2026
Type: conference-paper
Abstract: This study developed a machine learning framework for attendance forecasting and anomaly detection in educational access systems. The framework transformed student attendance, entry logs, exit logs and time information into structured attendance-access data for predictive analysis. Student absence minutes, absence frequency, entry-time regularity, exit-time count, previous absence status, and day-based patterns were coded into lateness minutes, absence frequency, entry-time regularity, exit-time count, previous absence status and day-based patterns, respectively, to capture student attendance patterns. Forecasting the attendance prediction model was developed and tested using random forest, XGBoost and LSTM models and the classification model was developed and tested with LSTM for attendance and isolation forest for detecting anomalies. Among the tested attendance forecasting models, XGBoost obtained the lowest root mean square error (RMSE = 0.007) and the highest coefficient of determination (R2= 0.086). The low absolute error indicates close short-term alignment with observed attendance rates; however, the modest R2shows that the available behavioral and temporal features explained only a limited portion of attendance-rate variability. For attendance-status classification, LSTM achieved the highest F1-score of 0.940. Isolation Forest achieved an accuracy of 0.920 but a recall of 0.591; therefore, it was interpreted as a conservative supplementary anomaly-screening component rather than a standalone anomaly detector. The framework converts these model outputs into risk indicators and alerts to support authorized school personnel in attendance monitoring and early intervention.
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: hybrid
OpenAlex ID: W7203466922