Detección de fraude en tarjetas de crédito mediante técnicas de minería de datos

Abstract

Data mining and machine learning are highly potential tools in the identi_cation of unusual observations in pattern trends, since they are a set of robust techniques that facilitate decision making, the knowledge discovery in databases process, KDD for its acronym in English, is a _eld of statistics and computer science that employs various techniques and methodologies for the process of identifying valuable patterns in the extraction of new, useful and novel information; One of the most important stages is the data mining, where the estimation of the parameters of the probabilistic models such as neural networks, random forest, naive bayes, vectorial support machines, generalized lineal models logit, probit and log log; later they will be evaluated and analyzed with the classi_cation, accuracy, recall, f beta score and roc curve metrics. Fraud is defined as the action contrary to the truth and rectitude, which harms the person or entity against whom it is committed, this leads to economic losses and legal problems. There are different types of fraud, as they are intrusive to private networks, credit card, telecommunications and money laundering. This work seeks to compare the efficiency of the probabilistic models of data mining, machine learning and generalized linear models to be applied to credit card transactions and evaluate with the classi_cation metrics that probabilistic model is efficient in predicting the fraud