Next-cell prediction with LSTM based on vehicle mobility for 5G mc-IoT slices

Abstract

Network slicing is one 5G network enabler that may be used to enhance the requirements of mission-critical Machine Type Communications (mcMTC) in critical IoT applications. But, in applications with high mobility support, the network slicing will also be influenced by users’ movement, which is necessary to handle the dynamicity of the system, especially for critical slices that require fast and reliable delivery from End to End (E2E). To fulfill the desired service quality (QoS) of critical slices due to their users’ movement. This paper presents mobility awareness for such types of applications through mobility prediction, in which the network can determine which cell the user is in near real-time. Furthermore, the proposed next-cell mobility prediction framework is developed as a multi-classification task, where we exploited Long Short-Term Memory (LSTM) and the collected historical mobility profiles of moving users to allow more accurate short- and long-term predictions of the candidate next-cell. Then, within the scope of high mobility mission-critical use cases, we evaluate the effectiveness of the proposed LSTM classifier in vehicular networks. We have used a real vehicle mobility dataset that is obtained from SUMO deployed in Bejaia, Algeria urban environment. Ultimately, we conducted a set of experiments on the classifier using datasets with various history lengths, and the results have validated the effectiveness of the performed predictions on short-term mobility prediction. Our experiments show that the proposed classifier performs better on longer history datasets. While compared to traditional Machine Learning (ML) algorithms used for classification, the proposed LSTM model outperformed ML methods with the best accurate prediction results.