Detección de fraude bancario mediante KDD y Aprendizaje Automático

Abstract

Abstract:This paper addresses the detection of banking fraud through the implementation of a knowledge discovery process in databases and machine learning algorithms. A starting point is a state-of-the-art study, examining the variables, tools, and metrics used. Based on this analysis, data available from public sources are selected and processed using cleansing, normalization, and feature selection methods. Experiments are carried out with supervised and unsupervised learning techniques, as well as deep learning techniques (neural networks). The results obtained show a performance of over 98%, demonstrating its ability to identify fraud patterns in the data used for experimentation. Finally, the selected model is implemented in a functional prototype developed with Python, creating a scenario that allows data to be processed in real time. The results obtained are compared with studies analyzed in the state of the art, which demonstrates significant progress in relation to evaluation metrics in an implementation scenario.