Machine Learning-Based Task Offloading in IoT-Fog Networks: A Comparative Analysis

Abstract

There is huge amount of growth of Internet of Things (IoT) devices which are generating and processing huge amounts of data. IoT devices are generally resource constrained so they can’t do all the computations properly. Fog Computing is emerged as a solution by putting the computation resources closer to the IoT devices which helps IoT devices to offload the heavy computational tasks to the nearby fog nodes. The main goal of research is to compare different ML methods used for deciding how the tasks will be offloaded in a IoT-Fog system. Proposed work have seven distinct approaches: three standalone machine learning models (Neural Networks, Random Forest, and Decision Trees), Deep Reinforcement Learning (DRL), a baseline reinforcement learning algorithm (RILTA), and two novel smart cascading hybrid models (Hybrid-NN-RF and Hybrid-NN-DT) that have been implemented and evaluated. Each algorithm has been thoroughly assessed using three performance metrics: offloading delay, energy consumption, and task allocation ratio. Experiments are done under different workloads ranging from 250 to 1000 IoT tasks to see the models performance as the number of IoT tasks increases. Overall, the work demonstrates how machine learning can improve task offloading efficiency in IoT–fog systems by reducing delay and energy consumption at scale.