Credit cards have become an integral part of the modern financial landscape, and their use is essential for individuals and businesses. This has resulted in a significant increase in their usage in recent years, especially with the growing popularity of online payments. Unfortunately, this increase in credit card use has also led to a corresponding rise in credit card fraud, posing a serious threat to financial security and privacy. Therefore, this research introduces a novel deep learning-based hybrid Harris hawks with sine cosine method for credit card fraud detection system (HASC-DLCCFD). The aim of the presented HASC-DLCCFD approach is to identify fraudulent credit card transactions. The suggested HASC-DLCCFD scheme introduces a HASC technique for feature selection, by combining Harris hawks optimization (HHO) with the sine cosine algorithm (SCA). For the purpose of identifying credit card fraud, an architecture of a convolutional neural network combined with long short-term memory (CNN–LSTM) is utilized in this study. Finally, the adaptive moment estimation (Adam) algorithm is utilized as a hyperparameter optimizer of the CNN-LSTM model. The performance of the suggested HASC-DLCCFD approach was experimentally evaluated using a publicly available database. The results demonstrate that the suggested HASC-DLCCFD approach outperforms other current techniques and achieved the highest accuracy of 99.5%.