Summary of results from 2024-25 earthquake prediction papers that use seismic statistical features.

Authors are listed alphabetically. ‘Split type’ refers to the splitting of the dataset into (train, validation [if used], test) and is one of {temporal, random, unknown}. ‘Model’ is the one for which the paper reported the best results. ‘Target 1’ is the condition under which the model output should be 1/‘yes’, as framed for the closest-given magnitude threshold to our M = 5. ‘Test performance low–high’ gives the lowest and highest test performance values reported (potentially depending on the time horizon D and geographical region). Abbreviations: DL = deep learning; ANN = artificial neural network.