Online Active Learning For LSTM-Based Network Traffic Prediction

Abstract

This paper introduces an online active deep learning framework tailored for intelligent and sustainable aerial-terrestrial IoT networks. The framework utilizes Long Short-Term Memory (LSTM) models, enhanced with active learning and incremental updates, to address the challenges of dynamic network traffic prediction. Model updates are selectively applied, prioritizing samples with higher prediction errors, enabling the system to adapt dynamically to evolving traffic patterns and improving predictive accuracy. The proposed approach accommodates non-stationary data streams and prioritizes critical information, establishing a robust foundation for effective network management. The framework's performance was validated using the Milan traffic dataset, demonstrating its effectiveness compared to non-weighted online learning methods and static models.