A weighted average ensemble learning based on the cuckoo search algorithm for fraud transactions detection

The identification and prevention of unauthorized credit card transactions are the goals of credit card fraud transaction detection. Cardholders and financial institutions can lose significant amounts of money with fraudulent transactions, which is why detection and prevention should be done as soon as possible. This paper suggests a new weighted average ensemble learning approach based on 20 decision trees (DT). The Cuckoo Search algorithm was used for searching for optimal weights. In this case, the ensemble learning model can be described as the weighted sum of predicted model probabilities, where each model’s probability is multiplied by its respective weight. For evaluation, many ensemble learning-based decision trees are used, including a decision tree, Random Forest (RF), Adaboost (AD), Gradient Boosting Tree Classifier (GB), and Extra tress (ET). To assess the effectiveness of the proposed solution, a randomly undersampled dataset of credit card transactions was employed. Several experiments are conducted based on a real dataset, which demonstrates a better prediction of the proposed ensemble learning approach in terms of accuracy (ACC), precision (PRE), recall score (REC), and F-measure (F).