Hierarchical hybrid neural network built from concatenating multiple nets after their feature extraction layers, creating a large fully connected layer resulting in a single binary classification for all inputs. Here we use as inputs the dedispersed frequency–time intensity array, frequency-collapsed pulse profile, DM–time array, and multibeam detection S/N, but our framework allows for any combination of these, as well as additional input data products and networks. Since the full, merged DNN is trained together, the model will learn the relative importance of each input. For example, the dedispersed intensity array will tend to have more predictive power than the multibeam statistics; thus, the former will have a greater influence on the output probability.