The burgeoning volume of credit card transactions, coupled with the increasing sophistication of fraudulent activities, necessitates the development of robust, intelligent fraud detection systems. This paper presents the development of a machine learning-based fraud detection system aimed at identifying and mitigating fraudulent credit card transactions in real-time. Traditional rule-based systems, while effective to a degree, often fail to adapt to the dynamic and evolving nature of fraudulent behaviours. Consequently, our approach leverages advanced machine learning techniques to enhance the accuracy, speed, and adaptability of fraud detection mechanisms. Feature engineering plays a critical role in our system, transforming raw transaction data into meaningful features that enhance the model’s ability to discern patterns indicative of fraud. Features such as transaction amount, time between transactions, geographical location, and merchant category are meticulously crafted and selected based on their predictive power. Additionally, anomaly detection methods are integrated to identify deviations from typical transaction behaviors, further bolstering the system’s detection capabilities. The proposed system employs a supervised learning paradigm, utilizing historical transaction data to train various machine learning models, including Logistic Regression, Decision Trees, Random Forests, and Gradient Boosting Machines. Each model is evaluated based on its precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC-ROC), to ensure a comprehensive assessment of performance. Special emphasis is placed on the imbalanced nature of the dataset, a common challenge in fraud detection, where fraudulent transactions constitute a minute fraction of the total transaction volume. To address this, we incorporate techniques such as Synthetic Minority Over-sampling Technique (SMOTE) and cost-sensitive learning to balance the dataset and mitigate bias. The system is designed to operate in a real-time environment, necessitating efficient data processing pipelines and model inference mechanisms. Moreover, the interpretability of machine learning models is paramount, particularly in financial contexts where transparency and accountability are critical. We employ techniques such as SHapley Additive exPlanations (SHAP) to provide insights into model decisions, facilitating trust and acceptance among stakeholders. This interpretability also aids in the continuous monitoring and refinement of the system, as it enables the identification of model drift and the incorporation of new fraud patterns as they emerge. Extensive empirical evaluation is conducted using a publicly available dataset of credit card transactions, demonstrating the efficacy of our approach. The results indicate that the machine learning-based system significantly outperforms traditional rule-based systems, achieving higher detection rates with lower false positive rates. Furthermore, the system exhibits robust performance across varying transaction volumes and fraud types, underscoring its versatility and scalability.