Predicting Fraud in Financial Payment Services through Optimized Hyper-Parameter-Tuned XGBoost Model

• Published in: Mathematics (Basel) Vol. 10; no. 24; p. 4679
• Main Authors: Dalal, Surjeet, Seth, Bijeta, Radulescu, Magdalena, Secara, Carmen, Tolea, Claudia
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.12.2022
• Subjects: Accuracy; Algorithms; Chargebacks; Datasets; Financial institutions; financial payment fraud; Food science; Fraud; fraud detection; Fraud prevention; hyper-parameter tuning; Identity theft; Machine learning; Mathematical models; nature-inspired optimization; Personal information; XGBoost; Romania
• ISSN: 2227-7390; 2227-7390
• DOI: 10.3390/math10244679

Online transactions, medical services, financial transactions, and banking all have their share of fraudulent activity. The annual revenue generated by fraud exceeds $1 trillion. Even while fraud is dangerous for organizations, it may be uncovered with the help of intelligent solutions such as rules engines and machine learning. In this research, we introduce a unique hybrid technique for identifying financial payment fraud by combining nature-inspired-based Hyperparameter tuning with several supervised classifier models, as implemented in a modified version of the XGBoost Algorithm. At the outset, we split out a sample of the full financial payment dataset to use as a test set. We use 70% of the data for training and 30% for testing. Records that are known to be illegitimate or fraudulent are predicted, while those that raise suspicion are further investigated using a number of machine learning algorithms. The models are trained and validated using the 10-fold cross-validation technique. Several tests using a dataset of actual financial transactions are used to demonstrate the effectiveness of the proposed approach.