Hyperparameter Optimization Using Genetic Algorithms to Detect Frauds Transactions

Abstract

Today, online services have evolved at a large scale which has made our life very easy, but there are many problems and challenges to make these services more secure for users. For instance, every day, many transactions are made by customers, and much private information is posted and shared on E-commerce and social media websites which makes privacy, safety and reliability a trough challenge to defy. Credit card fraud detection is one of these problems because fraudsters try to make every transaction legitimate by stealing the information related to the credit card. Hence, easy methods and other less complex techniques are not going to detect this type of fraud. Having an effective fraud detection technique has become a requirement for all banks to minimize chaos and maintain some order in place. In this paper, we use machine learning to detect fraudulent transactions by applying a genetic algorithm (GA) to optimize the hyperparameter and compare it with grid search (GS) methods. The used algorithms are random forest (RF), AdaBoost (AB), logistic regression (LR), decision tree (DT), and support vector machine (SVM) classifier. As the credit card fraud dataset is highly skewed (imbalanced data set) and the performance of fraud detection is greatly affected by the sampling approach, so we use undersampling to handle this issue. The obtained results in terms of accuracy, precision, recall, and F1_score have shown that the genetic algorithm can generate better performances in a short-time in comparison with the GS algorithm.