Credit card fraud detection is crucial for financial institutions to prevent unauthorized transactions; however, it is hindered by challenges such as high-dimensional data and class imbalance. This study proposes a novel approach that integrates the Binary Dragonfly Algorithm (BDA) for Feature Selection (FS) with K-Nearest Neighbors (K-NN) for classification. Applied to a credit card fraud dataset, the method achieves 99.14% accuracy, 98.52% recall, 99.78% precision, and 99.15% F1-score, outperforming existing techniques. This approach provides an effective solution for fraud detection, not only enhancing the precision of fraud detection but also optimizing the model's efficiency. Future work could explore combining BDA with other metaheuristic algorithms or advanced classifiers to enhance performance further.