Incorporating Feature Fusion Engineering and Ensemble Learning for Effective Credit Card Fraud Detection

Recent developments in electronic payments and ecommerce have led to a rise in financial fraud cases, including credit card fraud. Thus, it is imperative to implement systems that are capable of detecting fraudulent transactions with credit cards. To identify credit card scams using machine learning, we must carefully choose its subtle features. Researchers have developed numerous methods to identify this fraud. In contrast to previous research, we have developed a hybrid feature selection method in this paper that provides an improved approach for detecting credit card fraud. This study employs three machine learning (ML) models: extra tree (ET), extreme gradient boosting (XGBC), and random forest (RF). Preprocessing of the data, feature engineering (to select the best features), and ensemble voting comprise the three phases of this proposed work. While choosing features from both the filter and embedding methods, we used the optimal threshold values. In this paper, the Pearson correlation feature selection technique is applied to remove highly correlated features. Afterward, the remaining feature subset is passed through information gain and random forest importance techniques, and the most significant features are selected that are common in both methods. Thereafter, the classifiers undergo hyperparameter tuning to find the best set of parameters and are combined into an ensemble model using hard voting, enhancing prediction accuracy.