Privacy-Preserving On Screen Activity Tracking And Classification In E-Learning Using Federated Learning

Abstract

E-learning platforms provide flexible access to educational resources but also increase exposure to digital distractions and create privacy concerns when user activity is centrally monitored. This project develops a privacy-preserving on-screen activity tracking and classification system in which active applications and browser activity are processed locally and categorized as productive or non-productive. The implementation uses Flask for backend services, HTML/CSS/JavaScript for the dashboard, SQLite for local activity history, and machine-learning components including TensorFlow and scikit-learn. Federated Learning is used to support decentralized model improvement while the thesis states that raw activity information remains on the local device. The system calculates total screen time, productive learning time, non-productive time, productivity percentage, and application-wise usage statistics. Evaluation included unit, integration, system, functional, performance, usability, security, and compatibility testing. All 12 reported functional test cases passed, and the thesis reports successful operation of the user interface, activity tracking, classification, productivity analysis, database management, dashboard visualization, report generation, and Federated Learning modules. The model-accuracy graph shows classification accuracy increasing from 72% at epoch 1 to 96% at epoch 10. The resulting framework demonstrates a practical approach to productivity analytics for e-learning while maintaining local processing of sensitive activity data.