Optimizing Storage and Transformation in Edge Computing Through Interpretable Generalized Additive Neural Networks

Abstract

Edge computing is critical in promoting technologies like 5G, augmented reality, Internet of Things (IoT), and  vehicle-to-vehicle communication because it enables low-latency, portable, and location-aware services. This study focuses on a data transformation framework within edge computing environments. The preprocessing section uses derivative-free extended Kalman filtering (DF-EKF) to clean the input data once it has been first gathered from the BuptMecMigration-Edge-Computing-Dataset (BMECD). The processed data is subsequently fed into Interpretable Generalized Additive Neural Networks (IGANN) to manage data transformations and improve processing efficiency. Experimental results show a steady increase in average reward, reaching a plateau at approximately 0.22 after 4000 episodes, demonstrating effective training of the IGANN agent. Additionally, the average score increases from 57 to 59.5 after 1000 episodes, while the average task latency decreases from 1.6 seconds to 0.6 seconds. An examination of the correlation between task latency and task count reveals an initial rise in delay, followed by stabilization as the system optimizes resource allocation. These results show how effective edge computing is at task management and latency reduction in dynamic contexts.