Ensuring the integrity of attendance records remains a persistent challenge in academic environments, where proxy attendance and credential sharing undermine the reliability of conventional systems. While recent digital solutions have improved automation, they largely operate as passive record-keeping tools and fail to capture the behavioral context surrounding attendance events. This limitation creates an opportunity for misuse, particularly in web-based systems where authentication alone does not guarantee physical presence. To address this shortcoming, this study introduces a behavioral anomaly detection framework designed to identify proxy attendance through analysis of interaction patterns rather than explicit identity verification. The proposed system integrates a web-based attendance platform with a machine learning layer that models normal student behavior using metadata such as device fingerprints, submission timing, session concurrency, and geolocation traces. An Isolation Forest algorithm is employed to detect global anomalies within high-dimensional behavioral data, while a Local Outlier Factor (LOF) model provides secondary validation by examining local density deviations. The framework is evaluated using a synthesized dataset comprising 1,200 attendance records across multiple sessions, with controlled injection of proxy scenarios to simulate realistic misuse patterns. Experimental results indicate that the combined model achieves a precision of 87.3%, recall of 91.7%, and an AUC score of 0.943, demonstrating strong capability in distinguishing legitimate attendance from fraudulent activity with minimal latency overhead. The findings suggest that behavioral modeling offers a practical and scalable alternative to hardware-dependent biometric systems. This work contributes a deployable, hardware-independent framework that enhances attendance authenticity by embedding intelligence directly into web-based academic infrastructures.

Keywords: Behavioral Anomaly Detection, Proxy Attendance, Machine Learning, Isolation Forest, LOF, Web-Based System, Academic Integrity