Modelo de risco de fraude em transações de cartão de crédito no mercado de crédito bancário

Abstract

The economic, social and cultural developments that Brazilian society has gone through, especially in recent decades, were preponderant factors for the expansion of the use of credit cards as a means of payment, making them the most used today. Another important factor was the technological revolution, which introduced millions of new users to the digital world. These users began to use communication equipment, thus allowing purchase and sale transactions and the creation of a new digital market. In this way, the creation of new security mechanisms to avoid losses due to fraud has become a challenge for banking institutions, since the relationships between banks and customers have become increasingly digital. This monography makes a brief account of the development of the Brazilian credit card market, describing the main forms of fraud. A logistic regression model was applied using Kaggle's credit card fraud detection dataset with the purpose of predicting legitimate or fraudulent transactions. Two different ways of treating the data was proposed: the first, using the removal of correlated explanatory variables and the second, transforming the data through Principal Components Analysis, also known as PCA. The logistic regression model is applied to both forms of previous data treatment and their performance were evaluated. The model with the application of Principal Components Analysis presented better results for the study in question. This work is divided into ten parts, which are introduction, contextualization, problem definition, justification, objectives, data origin, data sampling, descriptive analysis, predictive model and final considerations.