Explainable Artificial Intelligence for Computation Offloading Optimization

Computation offloading has proven effective as a technology that enables mobile devices to run resource-intensive applications. Multi-access Edge Computing facilitates computation offloading for mobile devices. Compute-heavy tasks can be transferred from a mobile device to a nearby cloudlet to reduce computation time and to conserve the battery life of the mobile device. However, due to fluctuating network conditions and the limited computational capacity of the MEC nodes, the offloading decisions made by mobile devices might not always result in the lowest cost. This paper introduces a dynamic offloading framework for mobile users, taking into account the local overhead on the mobile device and the restricted communication and computation resources available on the network. We frame the offloading decision problem as a multi-label classification challenge and employ the eXtreme Gradient Boosting algorithm incorporated with Explainable Artificial Intelligence to minimize computation and offloading overhead. With this research, we show how to exploit Shapley Additive Explanations for feature selection, resulting in an optimal set of features leading to improved accuracy. Furthermore, the simulation results indicate that our approach can reduce system costs by up to 66.67% and 81.09% compared to the random offloading scheme and total offloading scheme, respectively.