Optimized Cache Placement in Edge Computing Using Machine Learning-Based Predictive Models

Abstract

Cache placement in edge computing is crucial in optimizing data retrieval efficiency, particularly in environments characterized by fluctuating content demand and constrained storage capacity. This study proposes a machine learning-driven approach to enhance cache placement decisions, employing Logistic Regression and Support Vector Machines as predictive models. To refine classification accuracy and computational efficiency, the models are optimized through gradient-based techniques such as Stochastic Gradient Descent, Mini-Batch Gradient Descent, Momentum Gradient Descent, Adam, and Newton's Method. The methodology involves processing real-world access data, where key features such as response time, access frequency, and file size inform caching decisions. Experimental evaluations reveal that Newton's Method delivers the highest classification accuracy 97.5% with rapid convergence, while Momentum Gradient Descent achieves the highest cache hit rate 13% at a cache size of 150 MB. Furthermore, the proposed machine learning models significantly surpass traditional caching strategies such as Least Recently Used and Least Frequently Used, achieving up to a 100% improvement in cache hit rate. These findings underscore the effectiveness of adaptive, data-driven caching techniques in enhancing network performance and reducing latency in real-time edge computing environments.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint