Data mining algorithms for risk detection in bank loans

This article proposes a new approach on detection of fraudulent credit operations applying computational intelligence techniques. We use a dataset of historical data of customers from a financial entity and we split it to train a classification and clustering algorithm. We train a radial basis function network to classify clients that commit or not credit fraud. Then, we build a Fuzzy c-means clustering to group data points to create customer profiles. This algorithm has the capacity of grouping the data inside clusters and assigning a degree of membership to the points outside the clusters. Subsequently, the trained classification algorithm is applied to the clusters to provide additional information about customer profiles. We demonstrate good performance for fraudulent credit operations and identification of customer profiles.