Detección de fraude financiero mediante redes neuronales de clasificación en un caso real español
This paper explores the possibilities offered by statistical tools based on artificial neural networks for pattern recognition in expert work for money-laundering detection. The data is provided by the Spanish Police Department and comes from a case in which is actually working at. Account information is provided, where some accounting entries are identified as fraud. Hence it is possible to use this information to train a classification model. In this analysis, after briefly describing methodology used and fitting strategy, it is presented a model with a promising predictive capacity, even with strongly unbalanced training data set. After applying balancing technique to the training data (SMOTE) the result is remarkably improved which would indicate the viability of those models as tool for police experts planification, providing a way to reduce the use of expensive research resources.