Despite their usefulness for volcano monitoring, emergent seismic signals, such as volcanic tremor or signals generated by lahars, are difficult to identify with confidence in a timely fashion. Machine-learning algorithms offer an objective alternative to traditional methods of identifying such volcanoseismic signals, because they are able to handle quickly large amounts of data, while requiring little input from the user. In this work, we combine permutation entropy and centroid as well as dominant frequency with supervised machine learning to evaluate their potential in identifying volcanic tremor and lahar signals recorded during the 2009 Redoubt volcano eruption. The particular dataset was chosen for the reason that the properties and occurrence times of the volcanoseismic signals during the eruption are well known from previous studies. We find that the selected features can effectively discriminate both types of signals against the seismic background, especially for stations that are near the source. Results show that the identification success rate for volcanic tremor reaches up to 96%, whereas this rate becomes up to 91% for lahar signals. The calculation of the features as well as the application of the machine-learning algorithms is fast, allowing their implementation in the operational environment of a volcano observatory during a volcanic crisis. Finally, the proposed methodology can potentially be used to objectively identify other emergent seismic signals such as tectonic tremor along subduction zones, glacial tremor, or seismic signals generated during landslides.