Aplicação de algoritmos e métodos estatísticos para a identificação de anomalias em transações financeiras

Abstract

With the growth of digital transactions and the emergence of sophisticated fraud, manual analysis has become unfeasible, requiring automated real-time systems. This study aims to apply data intelligence, combining statistics and machine learning, to identify financial anomalies to support decision-making. The methodology is divided into two main stages: first, methods such as Z-Score and Interquartile Range are employed for initial analysis and detection of extreme values. Complementarily, unsupervised learning is utilized with Density-Based Spatial Clustering of Applications with Noise, Local Outlier Factor, and Isolation Forest algorithms, allowing the mapping of suspicious behaviors in large volumes of data without the need for prior fraud history. The performance of the techniques was evaluated using integrated metrics and graphical representations. It is concluded that the integration between statistical modeling and data mining strengthens monitoring and process security, providing greater institutional protection against financial risks. The results indicate that the Isolation Forest achieved the best performance among the algorithms.