This paper reports a novel Machine Learning (ML) solution to power allocation problem modelled as a Stackelberg game. We consider a multi-relay cooperative environment where the performance of the relays is dependent upon the resources allocated to them. As a first step, a game theoretic framework has been used for modelling cooperation and competition among the relays. This Stackelberg game-based framework considers benefits of source and relays jointly utilizing a strategy for optimal power allocation based on incentivesprovided to the cooperating relays. Subsequently, an optimal set of relays is identified through machine leaning based bilevel optimization of objective functions which define the aforementioned Stackelberg game. The proposed ML optimization helps the source increase its utility by allocating optimal power to the participating relays at an optimal price. Results from simulation experiments confirm that the ML based solution of Stackelberg game optimization problem provides consistently better performance in terms of system throughput ascompared to the centralized scheme and an earlier reported heuristic scheme.