Effective Credit Card Fraud Detection Using Data Mining Techniques

In this chapter we gave an overview of the existing research, methodologies and techniques in credit card fraud detection that will pave the way to the methodological decisions presented in Chap. 3. The literature review first talked about the significance and challenges of fraud detection in the financial industry, and the difficulty of using the traditional rule-based systems to address this, and the rise in the use of machine learning methodologies to tackle this problem. Fraud detection research is heavily dependent on datasets; the review examined the features, strengths, and weaknesses of publicly available datasets especially the real-world dataset. Also included was discussion of the complexities of data pre-processing, such as handling missing values, encoding, categorical features, and class imbalances, given the importance of reliable techniques in preparing data for machine learning models. Besides, we also introduced various machine learning models used for fraud detection, including: traditional algorithms such as Logistic Regression; advanced ensemble methods such as Random Forest, XGBoost, LightGBM; and deep learning approaches such as DNNs. The strengths and limitations of each model were critically evaluated, from which the applicability of the models in solving fraud detection challenges was noticed. The capability of ensemble methods and deep learning models in dealing with high dimensional and imbalanced data was highlighted and traditional models were appreciated for their interpretability and simplicity. Importantly, it was found that feature engineering was a critical step in maximizing model performance, and gain based importance and cumulative feature importance were shown to intelligently reduce dimensionality while preserving predictive power. It also reviewed trade–offs between computational efficiency, interpretability, and performance as it pertained to the selection of feature engineering methods. Drawing insight from this literature review, the methodological choices in Chap. 3 are informed by best practices and employ the most up to date techniques. This chapter critically analyzes existing research to provide the theoretical and contextual basis required for developing an effective and efficient fraud detection system.