In massive multiple input multiple output (m-MIMO) technology, large number of antennas are used to boost the spectral efficiency (SE) and energy efficiency (EE) simultaneously. However, the design of such extensive and variable arrays is challenging and necessitates novel beamforming techniques to precisely direct signals toward intended users. In this work, the K-nearest neighbor (KNN) is fine-tuned using stochastic gradient descent (SGD) optimization, where the adaptability of KNN is used in beamforming, enhanced by the fast convergence properties of SGD. The hybrid beamforming method that combines the analog and digital beamforming is used, to control the phase the baseband signals processing. Beamforming algorithm is performed in a multi-cell (MC) framework, and the physical resource blocks (PRBs) at different base stations (BS) are scheduled to minimize required resources while maintaining the quality of service (QoS) for the mobile station (MS). This significantly enhances the SE and EE. The optimization process iteratively updates the beamforming vector, balancing SE and EE. The experimental results show that the proposed KNN-SGD achieves SE of 7.89 Mbits/J, EE of 8.97 Mbits/J, and total power consumption (PC) of 5.86 J better than the existing methods, deep neural network (DNN) and deep reinforcement learning (DRL).