Machine Learning-Empowered Mode Selection for QoS Improvement in D2D-Assisted Communication Networks

Device-to-device (D2D) communication has the potential to enhance network efficiency in upcoming sixthgeneration (6 G) wireless networks by enabling direct communication between users. However, in ultra-dense network environments, severe interference can degrade D2D performance, potentially leading to suboptimal Quality of Service (QoS) for some users. To address this challenge, we propose a machine learning-based approach for dynamic mode selection between D2D and cellular communication, aiming to improve network efficiency and ensure optimal user satisfaction. The proposed framework leverages clustering algorithms, including k-means and k-medoids, to segregate users based on channel gain, Signal-to-Interference-plus-Noise Ratio (SINR), and path loss. Furthermore, we employ binary classification models, including k-Nearest Neighbor (k-NN), Naive Bayes, and Support Vector Machine (SVM), to predict the optimal communication mode, using achievable data rate as the performance metric. Simulation results demonstrate that k medoids clustering coupled with a cubic SVM classifier achieves the highest accuracy of $98.36 \%$, leading to significant performance improvements. Moreover, k-medoids-based mode selection results in a $19.66 \%$ throughput increase and a $20.37 \%$ reduction in energy consumption, outperforming k -meansbased approaches for enhancing the QoS in D2D-assisted communication.