Deep Neural Networks-Based Optimization for Resource Allocation and Power Consumption in Networks Slicing

The rapid growth of communication networks and increasing demand for diverse services have intensified the need for efficient resource allocation and power-aware operation in 5G and beyond systems. Network slicing enables flexible sharing of physical infrastructure among heterogeneous services, but managing slice-level resources efficiently remains challenging. In this paper, a supervised deep neural network (DNN) regression framework is proposed to predict resource allocation decisions based on slice-level traffic and network features. Rather than employing reinforcement learning or explicit power modeling, the approach focuses on accurate allocation prediction, through which power efficiency is inferred by reducing resource over-provisioning. The mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE), and coefficient of determination (R2) are the metrics used to test the model. K-fold cross-validation is also used to check the efficiency of the model. Simulations are conducted in analogous conditions using standard machine-learning models such as support vector machine (SVM) regression and K-nearest neighbors (KNN). The proposed DNN demonstrates stable and precise allocation predictions, as it results moderate variance explanation (R2 = 0.6112) and minimal prediction errors (MSE = 0.0043, MAE = 0.0509, RMSE = 0.0659). The results show that supervised learning can be used to allocate resources in network slicing (NS) while taking power consumption into account.