Paper Title
REAL-TIME NETWORK JITTER SPIKE PREDICTION USING LSTM-BASED DEEP LEARNING

Abstract
Network jitter — the variation in packet arrival times — is a primary disruptor of real-time applications such as video conferencing, online gaming, and industrial IoT systems. Conventional threshold-based monitoring tools react to degradation only after it has occurred, leaving no room for proactive intervention. This paper presents a closed-loop, real-time jitter spike prediction system that combines a custom UDP measurement protocol with a two-layer Long Short-Term Memory (LSTM) deep learning model. The system continuously captures Round Trip Time (RTT), one-way delay, jitter, and packet-loss statistics, assembles them into sliding windows of twenty timesteps, and feeds each window to the trained LSTM to classify whether a jitter spike will occur within the next fifteen timesteps. On a held-out test set drawn from a locally collected dataset, the model achieves an accuracy of 88.44%, a recall of 1.000, and an F1-score of 0.4348. The perfect recall guarantees that no spike event goes undetected — a property of paramount importance for safety-critical network monitoring. An adaptive transmission controller adjusts the sender's inter-packet interval in real time in response to predictions, demonstrating the viability of end-to-end intelligent network management. Keywords - Network Jitter, LSTM, Real-Time Monitoring, UDP, Spike Prediction, Adaptive Control, Quality Of Service