Diagram of our approach to create a model that classifies radio sources as either ‘AGN remnant candidate’ or ‘non-candidate’ based on a number of morphological features. The solid arrows show the flow of our data, while the dashed arrows indicate the use of various trained models, specifically a SOM, HDBSCAN* clustered Haralick features, and a RF, all trained using only the data from the (undersampled) training set. The brackets indicate the different stages of our method and mention the sections in which corresponding details can be found.