Fraud Detection Through Nature-Inspired Algorithms

The proliferation of credit card payments has led to increased convenience but also an uptick in transaction fraud. This necessitates data-mining approaches to detect and prevent fraudulent activities. Transaction fraud detection involves multiple steps, including data analysis, preprocessing, feature selection, and hyperparameter optimization. Given the prevalence of anonymized features in transaction datasets, we focused on enhancing model performance through feature selection using nature-inspired algorithms. Our experiments revealed that this approach, when combined with machine learning models, holds promise for fraud prevention. Among the tested nature-inspired algorithms, the Grey Wolf Optimizer stood out, improving the model's ROC AUC score by 1.4% while selecting only half of the features compared to recursive feature elimination.