Optimization methods applied to neural networks for anomaly detection in credit card transactions

Abstract

In this project, it is intended to study optimization methods, especially stochastic algorithms, applied to the training of neural networks. In a recent paper proposed by BUENO, L. F. and MARTÍNES,J. M., "On the complexity of solving feasibility problems", 2018, a first-order algorithm was presented with good complexity results for least squares problems. One of this project's research main point is to develop a stochastic version of this algorithm. The performance of the studied algorithms will be analyzed when applied to fraud detection in credit card operations using Kaggle's Credit Card Fraud Detection database.