Credit card fraud causes substantial losses to credit card companies and consumers. Consequently, it is important to develop sophisticated and robust fraud detection techniques that can recognize the subtle differences between fraudulent and legitimate transactions. Current fraud detection techniques mainly operate at the transaction level or account level. However, neither strategy is foolproof against fraud, leaving room for alternative techniques and improvements to existing techniques. Transaction-level approaches typically involve the analysis and aggregation of previous transaction data to detect credit card fraud. However, these approaches usually consider all the transaction attributes to be equally important. The conditional weighted transaction aggregation technique described in this paper addresses this issue by leveraging supervised machine learning techniques to identify fraudulent transactions. Empirical comparisons with existing transaction level methods and other transaction aggregation based methods demonstrate the effectiveness of the proposed technique.