Illicit operation of credit cards is a critical issue for both financial institutions and consumers. Credit card companies constantly search for ways to minimize fraud while ensuring that legitimate transactions are not declined. Traditional fraud detection algorithms have quite high false decline rates, leading to many legitimate transactions being rejected. This paper investigates the causes of false declines and how historical transaction data can be leveraged to apply machine learning algorithmic techniques. A comprehensive credit card activity dataset is analyzed, utilizing a range of both supervised and unsupervised machine learning algorithms to train models aimed at identifying fraudulent transactions. The paper primarily focuses on reducing false declines without affecting false negatives. The results suggest that machine learning techniques can optimize the precision score of fraud identification mechanisms and minimize the incidence of false declines. An approach which can help credit card companies reduce the false declines rates while preserving high levels of fraud detection precision, ultimately improving the customer experience and increased revenue for credit card issuers.