Machine Learning-based Cache Optimization on MEC Platform

The amount of data generation is exponentially increasing over the past decade due to the widespread use of multimedia applications and social media platforms. Advanced real-time applications such as virtual reality, augmented reality, automated vehicles, smart homes, and intelligent traffic control systems have increased the demand for low latency. Many of these applications are delay-sensitive and put enormous stress on the core network to respond in real-time. CDN (Content Delivery Network) brings storage service to end-users proximity to provide low latency, high data throughput, and low traffic pressure to handle the problems mentioned above. Due to the limited storage capacity of the edge, only in-demand content should cache. Therefore, to optimally utilized the cache space, an efficient content caching and replacement policy is needed. To this end, in this paper, we propose an optimal content replacement algorithm. In this algorithm, a video request pattern is first generated based on a publicly available dataset. After that, a machine learning model is trained on cache logs data. As a result, the predicted video is deleted from the edge to make space for new videos. A real-time testbed is built on KOREN to check the performance of our model. The results based on MAE, MSE, and R-2 show that our model performs well in real-time scenarios.