Fraud detection in financial transactions is a critical area of research due to the growing scale and complexity of digital payments. Traditional supervised learning techniques often require labeled datasets, which are scarce and costly to obtain. This review explores recent advancements in machine learning approaches—particularly unsupervised and semi-supervised methods—for detecting fraudulent activities using unlabeled datasets. Techniques such as clustering, autoencoders, and anomaly detection models are analyzed for their effectiveness and adaptability. The paper aims to provide insights into current challenges, comparative performance, and future research directions in fraud detection without labeled data.