Drug discovery refers to the process of identification of specific-disease causing proteins and underscores the research efforts to derive a new medication that targets these proteins. As such the drug discovery process entails significant challenges as it is time consuming, data intensive, and involves an expensive developmental process which demands rigorous lab testing with high rates of uncertainty that the given drug will succeed. Therefore, it highlights the crucial need for machine learning methods to automate and hasten the drug discovery pipeline for improved healthcare and assist clinicians to make informed decisions for in-vitro testing. However, most real-world biomedical datasets suffer from statistical ill-conditioning issues such as the class imbalance problem where the fewer class of potential drug candidate protein conformations are overshadowed by the larger protein-pool of non-drug candidates. Hence, this leads to erroneous conclusions when machine learning techniques are directly employed for data-learning and classification purposes. Therefore, this work takes a revolutionary stance to counter the class imbalance problem through advanced machine learning techniques that maximize the prediction rate of potential drug candidate molecular conformations for the target proteins ADORA2A and OPRK1 and subsequently reduces the failure rates of the drug discovery process. Experimental evaluation of the proposed machine learning methodologies further substantiates the effectiveness of our approach for drug discovery process.