New drugs require a considerable amount of time and money from development to approval. For example, it may take nearly 15 years to produce a single drug, and some of them may cost as much as ${\$}$3 billion to develop. Meanwhile, the success rate of new drugs is very small. For instance, it is reported that, approximately 1% of all the drugs from initial development are approved. Therefore, realizing the feasibility of the futuristic approval of a new drug at an early development stage would greatly benefit drug marketers and users. In this paper, we develop a machine learning model to predict if a new drug will be approved or not, using the patent specifications for the drug filed in the early development stage. Specifically, we analyze each patent, focusing on product patents, via natural language processing. To achieve this, we use the abstract, claims, and description of the patent specification, as well as other information, including number of claims, inventors, and cited/citing references. Overall, our experimental results prove that we can predict drug approval with a high score of F1-score=0.944.