Machine Learning-Based Dynamic Queuing Model for Heterogeneous Traffic in Smart Home Networks

Abstract

Smart equipment in a house that allows remote automation, control, and monitoring a system that links and unifies differences is called smart home network. Devices can control and accessed via a central hub and can connect with one another, i.e., smart phone app or voice assistant. Sensors, smart appliances, and other gadgets that can interact with one another and react to user or system provider commands are the elements that make up a smart home network. In order to accommodate a high number of smart devices with different distributions and heterogeneous physical access (wired and wireless), smart home networks are expanding quickly. This leads to a variety of traffic patterns. To classify heterogeneous traffic flows in smart home networks by developing a dynamic queuing model based on machine learning and utilizing the Semi-Supervised Support Vector Machine (S3VM) is the goal of the work. We evaluate the performance of the QoS-level Pair Heterogeneous-sourced traffic model (QP-SH) against that of the proposed dynamic queuing system. Higher priority traffic might not need a smaller delay than lower priority traffic, because many advance QoS-aware scheduling techniques merely use the traditional IP type of service (ToS) information to determine priority metrics for deciding how to divide bandwidth. For instance, traffic from streaming devices needs a shorter maximum latency than traffic from medical sensors, although the former has a higher priority. Proposed network was evaluated using the MATLAB simulator, which was used to conduct simulation tests to verify the suggested model's performance. The MAL-DQ algorithm was compared with an existing queuing model, QP-SH, across metrics such as classification accuracy, throughput, and delay. The results shows that MAL-DQ outperformed QP-SH by achieving higher accuracy (average of 90.57% vs. 82.27%), better throughput (8.95 kbps vs. 7.02 kbps), and lower average delay (6.62 ms vs. 15.98 ms), confirming its suitability for smart home environments.