In the last decades seismic hazard studies have improved since larger and richer geological databases are available. These databases allow the analysis of seismic behavior of active faults. The purpose of these studies is to detect new sources of risk, generally unknown, named seismotectonic provinces. These are regions where the existence of seismic risk is known but faults are unknown. The method widely used to detect seismotectonic provinces is based on visual, non-automatic inspection of seismic events. This approach employs geological and tectonic zoning maps collected in seismic catalogs. However, it must also be recognized that the procedure described involves great uncertainty in identifying seismotectonic regions. The reason is that it is based on a visual, qualitative and subjective analysis of data. Although human vision is a powerful sense it is well known that studies based on it are prompt to lack repetitiveness and are tediousness in cases such as the manual analysis of large visual databases.

An efficient alternative to cope with the manual analysis of large seismological data collections is spatial data mining. The aim of spatial clustering of seismic data is to group seismic events into clusters in such a way that each cluster represents a seismotectonic province. The purpose of this work is to determine seismotectonic provinces in South West Colombia by using spatial data mining techniques. For this, an initial assessment of the South-West Colombian Earthquakes Catalogue’s (CATOCC) internal structure is made through a spatial randomness test (Hopkins Test) in order to determine if the data comprise aggregated structures. Afterwards, two clustering algorithm implementations are made: the first one with a Hard approach (k-means) and the second with a Fuzzy focus (fuzzy k-means) in order to identify groups that may correspond to seismotectonic provinces. Finally, a validation of the obtained partitions is made through relative validation indexes (Generalized Dunn index, Xie and Beni and the fuzzy hypervolume) to verify if the resulting groups represent an optimal partition.