Learning latent representations of bank customers with the Variational Autoencoder

Learning data representations that reflect the customers’ creditworthiness can improve marketing campaigns, customer relationship management, and process management or credit risk assessment in retail banks. In this research, we show it is possible to steer latent space of Variational Autoencoder (VAE) using a semi-supervised learning framework specific grouping input called Weight Evidence (WoE). Our proposed method learns representation showing well-defied clustering structure. The structure captures creditworthiness, which unknown priori cannot be identified space. main advantages our are natural data, suggests number clusters, spatial coherence generates unseen customers assign them one existing clusters. empirical results, based on real sets reflecting different market economic conditions, none well-known models benchmark analysis able obtain well-defined structures like method. Further, how banks use methodology campaigns assessment.