We systematically and comprehensively tested almost 100 feature groups in four commonly used automatic event classifiers to find the combinations that maximize the classification performance for long-period and volcano-tectonic seismic events at Cotopaxi volcano, Ecuador. The feature groups tested fall into the following categories: time, fast-Fourier, wavelet transform, intensity-statistics, shape, and texture. An analysis of the relevance of feature groups highlights and finds that each classifier performs similar to one another when using the best combination of features, as determined by the mean of the area under the curve metric, with no statistically significant differences between them. The feed-forward back-propagation neural network and random forest classifiers achieved scores of 0.96 while the naive Bayes and k-nearest neighbor (kNN) classifiers achieved scores of 0.95. The shape and texture feature groups were the most frequently utilized (three times each) among the different classifiers, and thus, the most appropriate for classifying both types of seismic events. The kNN ( k=3 ) classifier using the shape feature group reached an excellent performance with lower algorithmic complexity. Moreover, nontraditional features like those computed from spectrogram images overcame traditional time, Fourier, and scale features when classifying seismic events.