The prediction of earthquakes is a challenging task that has been largely addressed with not very satisfactory results so far. In this work, the historical data of one Chilean region -one of the countries with higher seismic activity- is analyzed. First, raw data are processed and aftershocks and foreshocks are removed in order to predict only main shocks. New attributes are also generated. Then, a clustering algorithm is applied to these data according to several suggestions found in the literature. This way the data forming the time series are labeled and, somehow, discretized. Finally, sequence of labels acting as precursory patterns are searched for within the dataset. The results achieved lead to conclude that the method is promising and could be applied to other active regions.