Fraud detection in financial transactions continues to be a major challenge in the digital age, primarily due to the ever-evolving tactics used by fraudsters, which create substantial risks for financial institutions and their clients. Traditional detection methods often face challenges in issues such as data imbalance, high false positive rates, limited adaptability to new fraud strategies, and a lack of interpretability, all of which reduce their effectiveness in practical applications. Additionally, the inherent uncertainty in detecting fraudulent activities, due to the unpredictability of fraud patterns and the evolving nature of fraud techniques, further complicates accurate detection and decision-making. To address these issues, this paper proposes a novel multilayered fraud detection framework that integrates a Decision Tree (DT) classifier with a fuzzy inference system (FIS) to enhance both accuracy and transparency. The first layer involves the DT classifier, which learns decision rules from the input data—preprocessed through SMOTE for class balancing and Min-Max normalization—and produces an initial classification based on crisp decision boundaries. These learned rules are then translated into fuzzy rules to construct the second layer, the FIS, which evaluates transaction data using degrees of certainty, allowing it to handle ambiguous or borderline cases more effectively. Finally, a hard-voting mechanism acts as the third layer, aggregating the outputs of the DT and FIS to make a robust final decision. This multilayered strategy enables the system to combine the precision of rule-based classification with the flexibility of fuzzy reasoning, leading to improved detection of complex fraud patterns. Experimental validation on real-world financial transaction data demonstrates that the proposed approach achieves an outstanding accuracy of 95.98%, outperforming traditional models like Random Forest, CNN-SVM, and standalone rule-based systems.