Computational Offloading and Resource Allocation for IoT applications using Decision Tree based Reinforcement Learning

The pervasive penetration of IoT devices in various domains such as autonomous vehicles, supply chain management, video surveillance, healthcare, industrial automation etc. necessitates for advanced computing paradigms to achieve real time response delivery. Edge computing offers prompt service response via its competent decentralized platform for catering disseminate workload, hence serving as front-runner for competently handling a wide spectrum of IoT applications. However, optimal distribution of workload in the form of incoming tasks to appropriate destinations remains a challenging issue due to multiple factors such as dynamic offloading decision, optimal resource allocation, heterogeneity of devices, unbalanced workload etc in collaborative Cloud-Edge layered architecture. Employing advanced Artificial Intelligence (AI)-based techniques, provides promising solutions to address the complex task assignment problem. However, existing solutions encounter significant challenges, including prolonged convergence time, extended learning periods for agents and inability to adapt to a stochastic environment. Hence, our work aims to design a unified framework for performing computational offloading and resource allocation in diverse IoT applications using Decision Tree Empowered Reinforcement Learning (DTRL) technique. The proposed work formulates the optimization problem for offloading decisions at runtime and allocates the optimal resources for incoming tasks to improve the Quality-of-Service parameters (QoS). The computational results conducted over a simulation environment proved that the proposed approach has the high convergence ability, exploration and exploitation capability and outperforms the existing state-of-the-art approaches in terms of delay, energy consumption, waiting time, task acceptance ratio and service cost.