The growth of Internet of Things (IoT) devices continues to experience an exponential rise due to their vast applications in various industries. However, sustaining their operational lifetime is a major challenge due to critical factors such as the need for frequent recharging of energy buffers. The recent convergence of green energy harvesting (EH) and IoT technologies has proven to be a potential solution to tackle this challenge. However, the intermittent characteristics of green energy sources and wireless fading channels pose another challenge, as the quality of service in IoT networks revolves around the available energy and channel conditions. The traditional optimization strategies based on non-causal knowledge about these random quantities are deemed unsuitable for the realization of an autonomous IoT network operation. To combat this challenge, various algorithms from the field of machine learning (ML) have been proposed as potential solutions. Therefore, in this article, we aim to investigate the applications of ML algorithms in EH IoT networks. To achieve this, first, we provide an overview of ML categories commonly adopted in IoT networks. Secondly, due to the peculiarity of EH IoT networks, we provide an extensive description of the MLcategories widely explored in this domain. This includes reinforcement learning, deep learning, and deep reinforcement learning. Thirdly, we present a review of studies where the applications of the aforementioned MLalgorithms are demonstrated. Further, we identify challenges that are likely to impact the implementation of these algorithms. In conclusion, we highlight unexplored and emerging areas for potential future research.

In this paper, we exploit the spatial and transmission diversities in cooperative non-orthogonal multiple access (C-NOMA) networks to improve the system sum-rate. To achieve this, we propose a user-pairing scheme where near-field user pairs serve as relays for user pairs in the far-field region. Based on this pairing scheme, we incorporate a space-time block code transmission technique at the near-field user pairs to maximize the transmission diversity in the cooperative phase. Moreover, we consider a non-linear energy harvesting model at the near-field user pair to alleviate the problem of energy consumption during the cooperative transmission phase. Further to this, we formulate a sum-rate maximization problem that is addressed from the viewpoint of joint power allocation factor and power splitting ratio optimization. We develop a low-computational iterative algorithm based on the concepts of the Stackelberg game and the Nash bargaining solution. We benchmark our findings with different scenarios, such as energy harvesting C-NOMA with a fixed power allocation factor and power splitting ratio, energy harvesting C-NOMA without STBC, non-cooperative NOMA, and orthogonal multiple access. The results obtained via simulations outperform the benchmark schemes.