This algorithm exploits the relationships between variables to improve the reconstruction performance of the variational autoencoder (VAE). A correlation score was used as the metric to group the features via a distance-based clustering method. The resulting clusters served as inputs for the Attention-Based VAE.