Utilization of machine learning in future wireless networks for resource optimization: A survey

Future wireless networks will play an essential role as the need for performance and feature availability grows. Most of the traffic in future wireless networks is due to increased Internet of things (IoT) devices, making resource optimization critical. Traditional optimization algorithms have limitations due to their high computational complexity, which restricts their use in modern applications. To address this, machine learning algorithms are now the preferred alternative to traditional optimization algorithms due to their improved runtime complexity. We present a comprehensive survey on the use of machine learning for resource optimization in future wireless networks. The use of machine learning is divided into three categories: (i) comprehensive solutions, where machine learning is the primary component of the solution approach; (ii) partial solutions, where machine learning is used alongside a traditional approach for optimization; and (iii) environment-only solutions, where optimization is performed in a machine-learning environment. We have further classified objective functions (e.g., energy, latency, data rate, etc.) within each category based on the pure objective function, variations on the objective function, and objective function tradeoffs with respect to other objective functions. We present objective functions and constraints used in the literature for optimization problem formulation. We provide an overview of frequently used machine learning algorithms for resource optimization, followed by a detailed survey of machine learning works in the literature in the three aforementioned categories. Finally, we discuss future research directions for utilizing machine learning to optimize resource management in future wireless networks.