Early detection of seismic events is fundamental to mitigate the damages produced by these natural phenomena, and mainly, save lives. In this paper, logistic regression models were applied to distinguish between seismic and non-seismic local events considering data segments from four seismological stations (epicenter in the Santander department, Colombia) of the National Seismological Network of Colombia. The classification outputs obtained by each individual classifier were then combined using voting functions. For each station, Degree of Polarization, Ratio of Vertical power to Total power, Skewness and Kurtosis of the three-component seismic data were extracted. Results showed that the best individual classifier achieved an accuracy of 95% and reaches an accuracy of 98% using voting functions. This classifier is the first step in the development of alert mechanisms for the early detection of seismic events in Colombia.