The present disclosure relates to determining detecting blast electronic activities. A method can include identifying a plurality of first electronic activities transmitted by a first electronic account of a data source provider. For each first electronic activity of the plurality of first electronic activities, a plurality of features can be extracted. For at least one first electronic activity of the plurality of first electronic activities, a blast probability score can be generated indicating a likelihood that the at least one first electronic activity is a blast electronic activity. The blast probability score can be generated using a machine learning model trained using features extracted from second electronic activities labeled as blast electronic activities and features extracted from third electronic activities labeled as non-blast electronic activities. An association between the at least one first electronic activity and the blast probability score can be stored in a data structure.