Abstract: Load balancing of 5G cellular networks is achieved by reducing network congestion utilizing two components of learning and optimization. First, a number of learning approaches including Linear Least Square Regression (LLSR), Auto Regressive Integrated Moving Average (ARIMA), and Multi-Layer Perceptron Deep Learning (MLPDL) are used to model either Physical Resource Block (PRB) or Packet Dedicated Control CHannel (PDCCH) utilization as a function of average connected user equipment and predict the number of average users corresponding to predefined thresholds of congestion in utilizing cellular towers. Then, an optimization problem is formulated to minimize 5G network congestion subject to constraints of user quality and load preservation. Three alternative solutions, namely Constrained Simulated Annealing (CSA), Block Coordinated Descent Simulated Annealing (BCDSA), and Genetic Algorithms (GA) are presented to solve the optimization problem.