Lightweight Latency Prediction Scheme for Edge Applications: A Rational Modelling Approach

2025-11-04

AI Summary

To address the challenge of accurately predicting end-to-end network latency in real-time edge computing—thereby hindering reliable task offloading—this paper proposes a lightweight, non-intrusive latency prediction model. Unlike existing approaches relying on active probing or complex neural networks, our method introduces a novel rational-function-based modeling paradigm, integrating low-overhead features such as frame size, arrival rate, and link utilization to achieve both high accuracy and ultra-low inference latency. Through rigorous feature engineering and five-fold cross-validation, the model achieves MAE = 0.0115 and R² = 0.9847 on real-world edge deployments—significantly outperforming conventional regression and lightweight neural network baselines. The proposed framework offers a new paradigm for resource-constrained edge nodes, balancing predictive accuracy, computational efficiency, and practical deployability.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Planning, Routing, and Scheduling: Model-Based Reasoning
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Accurately predicting end-to-end network latency is essential for enabling reliable task offloading in real-time edge computing applications. This paper introduces a lightweight latency prediction scheme based on rational modelling that uses features such as frame size, arrival rate, and link utilization, eliminating the need for intrusive active probing. The model achieves state-of-the-art prediction accuracy through extensive experiments and 5-fold cross-validation (MAE = 0.0115, R$^2$ = 0.9847) with competitive inference time, offering a substantial trade-off between precision and efficiency compared to traditional regressors and neural networks.

Problem

• Predicting network latency for reliable edge task offloading
• Developing lightweight model using frame and link features
• Achieving high accuracy with efficient computational performance

Innovation

• Lightweight latency prediction using rational modeling
• Eliminates intrusive probing with frame features
• Achieves high accuracy with competitive inference time