Credit card fraud remains a persistent and evolving challenge for banks and financial service providers, leading to substantial financial losses and erosion of customer trust. This paper presents an adaptive machine learning-based framework for credit card fraud detection that learns transaction-level behavioral patterns to effectively identify anomalous activities. The proposed approach follows a complete analytical pipeline, including data cleaning, normalization, feature engineering, and robust class imbalance handling through a hybrid strategy combining SMOTE oversampling and random undersampling. An XGBoost (XGBClassifier) model is employed due to its ability to handle large-scale, imbalanced data and capture complex non-linear relationships, supported by systematic hyperparameter tuning to improve generalization. Model performance is evaluated using precision, recall, F1-score, and accuracy to provide a balanced assessment of fraud and nonfraud predictions. Experimental results on a large dataset demonstrate improved detection performance with reduced false positives and false negatives. Overall, the study highlights the effectiveness of adaptive, scalable, and real-time machine learning solutions for modern credit card fraud detection systems.