Abstract
This project addresses the critical issue of fraud detection in credit card transactions, an imperative conand cern for both financial institutions and cardholders. With the increasing sophistication of fraudulent activities, accurate identification and prevention of fraudulent transactions have become paramount. The study focuses on a dataset comprising credit card transactions conducted by European cardholders in September 2013. Notably, the dataset exhibits a severe class imbalance, with fraudulent transactions accounting for a mere 0.172% of the total. The primary objective of this research is to develop a robust machine-learning model capable of effectively discerning between legitimate and fraudulent transactions. The project commences with an extensive exploration of the dataset, encompassing checks for data imbalance, feature visualization, and analysis of feature interrelationships. Subsequently, four predictive models, including Random Forest, AdaBoost, Cat Boost, and XG Boost, were employed and evaluated. The dataset was partitioned into three subsets: a training set, a validation set, and a test set. Initial results showcased promising performance, with the Random Forest model yielding an Area Under the Curve (AUC) the core of 0.85 on the test set. The AdaBoost model achieved a slightly lower AUC score of 0.83, while the Cat Boost model, following 500 iterations, attained an AUC score of 0.86. The XG Boost model demonstrated exceptional promise, achieving a validation score of 0.984, and subsequently producing an AUC score of 0.974 on the test set. Further, the project introduced a Light GBM model, leveraging both train-validation split and cross-validation methods. The former yielded AUC scores of approximately 0.974 on the validation set and 0.946 on the test set. Cross-validation exhibited a similar effectiveness, culminating in an AUC score of 0.93 on the test predictions. This study not only underscores the efficacy of employing advanced machine learning techniques in fraud detection but also emphasizes the importance of model selection and evaluation in the context of imbalanced data. The findings provide valuable insights for financial institutions seeking to bolster their fraud detection capabilities, ultimately enhancing the security and trust of credit card transactions.