Multi-Step-Ahead Prediction of User Movement on 5G Networks Using the Long Short-Term Memory Method

Since the initial launch of 5G networks in 2018, 5G technology has rapidly expanded to cover about 40% of the world's population, offering higher bandwidth, faster connectivity, and lower latency than 4G. However, due to its higher frequency operation, 5G necessitates more frequent handovers, making efficient handover performance crucial. This research focuses on enhancing handover efficiency by predicting user movement using Long Short-Term Memory (LSTM) models for multi-step-ahead prediction. The model created will utilize 60 seconds of historical data to predict 10 seconds ahead, with a target to achieve an average prediction accuracy of less than 20 meters with a latency under 100 ms. The study employs TensorFlow for model training, leveraging the Grab-Posisi dataset, and utilizes Plotly and TensorBoard for result visualization. Six model variations were trained, with the 16-unit model trained for 3106 epochs performing best, achieving 16.99 meters accuracy and 31 ms latency. The findings suggest that LSTM is effective for predicting user movement in 5G networks, with potential accuracy improvements through extended training epochs.