Least recently used (LRU) caching policy based on naïve bayes ML algorithm in collaborative peer-to-peer systems for network bandwidth utilization

Web caching offers several advantages, such as increasing cache hit rates, lowering the workload on origin servers, and minimizing network traffic. Nevertheless, limited cache capacity poses a major challenge in web caching systems. Moreover, repeatedly fetching same media objects from origin servers leads to unnecessary bandwidth consumption. Furthermore, traditional caching policies, including Least Recently Used (LRU), are vulnerable to cache pollution. This study introduces a collaborative caching policy based on the Naïve Bayes (NB) Machine Learning (ML) algorithm. The proposed policy exploits structured peer-to-peer architectures, allowing cache contents to be shared among peers to improve the efficiency of LRU web caching policy. Performance evaluation is conducted through simulations using two real-world datasets obtained from YemenNet Internet Service Provider (ISP) and the IRCache network. The results show that the proposed policy outperforms the traditional LRU policy in terms of Hit Ratio (HR), Byte Hit Ratio (BHR), and Cost Throughput (CT). For example, in some cases the Improvement Ratio (IR) of is more than 12% for YemeNet dataset; while it is more than 24% for IRCache dataset.