Fraud Detection of credit card fraud in the financial institution is one of the most important issues because fraudulent transactions are low in number, datasets are asymmetric and credit card transaction features are high dimensional. In this paper, I suggest a hybrid architecture with Principal Component Analysis (PCA) to reduce the magnitude of the data and Isolation Forest (iForest) to identify anomalies. Evaluation was done using the Kaggle Credit Card Fraud dataset that contained 284,807 transactions with 492 fraud cases. PCA transformed the 30 original features into 12 significant components capturing 95 percent of the variance, and decreasing the number of computations. Then, Isolation Forest was used to identify outliers, using some parameters that were paramount to the imbalance of the dataset. The evaluation of the results of the experiment revealed that the suggested PCA+iForest framework demonstrated the F1-score of 0.92, which surpasses the Isolation Forest itself and the similar deep learning models without compromising the computational efficiency. The results show the advantages of the hybrid method in the process of detecting frauds, which provides a compromise between accuracy and scalability.