A Comparative Study of Traditional and Automated ML Models for Credit Card Fraud Identification

In a modern era, fraudsters can easily commit fraud with the help of advanced technology. Credit card fraud is the most prevalent forms of fraud. Fraudsters steal credit card details through different methods therefore, there is a need for machine learning model to track fraudulent transactions and prevent loss. This research focuses on the application of Auto ML in detecting fraudulent in credit card transactions and evaluating its performance against various ML classifier algorithms. Due to the amount of imbalanced data, machine learning techniques struggle to reduce misclassification costs while simultaneously detecting fraud with good prediction accuracy. Therefore, this research integrates sophisticated sampling techniques such as oversampling using SMOTE and undersampling using Random Under Sampler which solve class imbalance issue. The results highlight Auto ML's superior accuracy and reliability with Accuracy of 0.99, AUC of 0.98, Recall of 0.83 and Training Time of 602s and Prediction Time of 0.289s.