Lightweight Deep Learning Model for Latency Prediction

This technical report presents a lightweight Deep Learning (DL) model designed for predicting Round Trip Time (RTT) in 5G and beyond networks. The objective of the study is to enhance Quality of Service (QoS) for end-users by optimizing latency, a critical parameter for applications such as autonomous driving and real-time services. The authors utilize a novel dataset collected through a smartphone-based methodology in real-world 5G scenarios. The report details the experimental setup, including the selection of a lightweight Multi-Layer Perceptron (MLP) model, which is compared against more resource-intensive models. The findings indicate that the lightweight MLP model outperforms traditional models in terms of Mean Squared Error (MSE), demonstrating its suitability for resource-constrained environments at the network edge. Additionally, the report discusses the use of the Shapley technique to improve the interpretability of the model's predictions, thereby promoting its adoption in practical applications. The document concludes with suggestions for future research directions.