The existing literature on device-to-device (D2D) architecture suffers from a dearth of analysis under imperfect channel conditions. There is a need for rigorous analyses on the policy improvement and evaluation of network performance. Accordingly, a two-stage transmit power control approach (named QSPCA) is proposed: First, a reinforcement Q-learning based power control technique and; second, a supervised learning based support vector machine (SVM) model. This model replaces the unified communication model of the conventional D2D setup with a distributed one, thereby requiring lower resources, such as D2D throughput, transmit power, and signal-to-interference-plus-noise ratio as compared to existing algorithms. Results confirm that the QSPCA technique is better than existing models by at least 15.31% and 19.5% in terms of throughput as compared to SVM and Q-learning techniques, respectively. The customizability of the QSPCA technique opens up multiple avenues and industrial communication technologies in 5G networks, such as factory automation.