Reinforcement learning has recently gained significant attention in the domain of financial trading due to its success in sequential decision-making problems as in the Forex market, for example. The intricate nature of learning achieved by this class of algorithms, however, often obscures the rationale behind their decision-making processes, raising concerns of opacity and unpredictability. These concerns hinder their application in high-stake domains. Explainable artificial intelligence is a relatively recent field that has emerged to overcome these shortcomings and is concerned with techniques that facilitate interpretability of the results returned by black-box models. In this vein, a novel optimisation procedure is proposed in this paper, which leverages Shapley values and the resulting increase in model interpretability, to select the most appropriate technical market indicators aimed at increasing algorithmic trading performance on the Forex market, based on deep reinforcement learning. The utility of this approach is demonstrated in the form of a case study which involves the training and evaluation of two deep reinforcement learning agents, namely a Deep Q-network agent, and a Proximal Policy Optimisation agent.