Deep Cross Network-TabNet Hybrid Model for Intelligent Task Offloading Decision-Making in Mobile Edge Computing

Read full text Summarise with VIRAT
Authors: Shoukath Cherukat, J. Benita
Cited by: 0
References: 28

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Mobile Edge Computing (MEC) plays a key role in supporting latency-sensitive Internet of Things (IoT) services by enabling computation to be performed closer to end devices.However, accurate task offloading, which involves deciding whether a task should be executed locally on the mobile device or offloaded to an edge cloud remains challenging due to rapidly varying network quality, fluctuating traffic demand, heterogeneous access technologies and time-varying edge resource availability.Conventional rule-based and simplified optimization strategies struggle to adapt to such environments and often fail to capture complex, nonlinear dependencies among system variables.To address these challenges, this study proposes a hybrid Deep Cross Network-TabNet framework for intelligent task offloading decision-making, formulated as a supervised binary classification problem that predicts the execution mode (Edge Cloud vs. Mobile Device) from heterogeneous tabular inputs.The model fuses two complementary learners: a Deep & Cross Network (DCN) branch that explicitly models bounded-order feature interactions using three cross layers alongside a deep multilayer perceptron and a TabNet branch that performs instance-wise sparse feature selection through sequential attention masks across five decision steps with a 32-dimensional feature transformer.Representations from both branches are concatenated and passed to a fusion head with a sigmoid output.Experiments on a publicly available dataset demonstrate excellent predictive performance, achieving 0.986 test accuracy with weighted precision, recall and F1 score of 0.986 and ROC-AUC of 0.994.Comparative evaluation against strong tabular baselines confirms the incremental advantage of the proposed hybrid architecture.Beyond classification metrics, an objective-based evaluation using a principled latency-energy cost model shows that the proposed approach reduces the cost gap relative to a derived cost-optimal policy compared with the observed execution behavior.Ablation confirms the hybrid advantage over standalone baselines and stratified 5-fold cross-validation reports 0.986 ± 0.001 mean accuracy, indicating stable generalization within the evaluated setting.

Fields of Study & Subjects

IoT and Edge/Fog Computing Energy Efficiency in Computing Internet of Things and AI

Publication details

Journal / source
International journal of intelligent engineering and systems
Volume
19
Issue
4
Pages
1097–1121
Publication date
23 March 2026
Language
EN