Banca de QUALIFICAÇÃO: Alex Cerqueira Pinto

Fraud and Anomaly Detection Models in a National Bank: Theoretical Perspective and Empirical Application

Fraud Detection; Anomalies; Bibliometrics; Machine Learning; Banks

Detecting fraud and anomalies in financial companies is a critical concern to ensure the security and integrity of financial transactions. As the complexity of fraudulent activities increases, advances in detection models have become essential to identify and mitigate threats. This work seeks, firstly, to analyze and verify existing connections in the state of the art literature on anomaly and fraud detection in banks and to propose two practical applications in the development of empirical models using data from a large national bank. To this end, this work is segmented into three articles, represented in chapters four to four, and chapters two, presenting a systematic review of the literature on the topic and applied bibliometrics and complex network statistics to identify the citation connections that exist between them. and raise the most relevant considerations and challenges in the literature, such as: class imbalance; The need for real-time detection; Interpretability and evolution of our fraud patterns with the increase in fraud cases linked to social engineering. Based on these results, in the third chapter, a case study is proposed for the development and empirical application of the anomaly detection model, based on data from a Brazilian financial institution, in security products for individuals and legal entities. With the aim of complementing the collaboration with the literature and deepening the analysis on the topic, in the fourth chapter, it is proposed to develop and empirically apply a multi-class analytical model of the probability of fraud in credit operations and verify the performance of these models to mitigate fraud. fraud, including operations involving social engineering. These studies are useful for the scientific literature that studies financial risk, as well as for the financial industry and professionals responsible for detecting fraud, and provide a space to propose alternatives to solve these gaps and needs of the segment. One of the main contributions of this work is to improve the main techniques, apply the models empirically, debate the results and especially the possibility of generating innovation or proposing a possible solution to the specific challenge of fraud through social engineering and opening the possibility of applying two models originating from scores in different circumstances in the origination of credit operations