Data mining methods for gis analysis of seismic vulnerability

This paper aims at designing some data mining methods of evaluating the seismic vulnerability of regions in the built infrastructure. A supervised clustering methodology is employed, based on k-nearest neighbor graphs. Unlike other classification algorithms, the method has the advantage of taking into account any distribution of training instances and also data topology. For the particular problem of seismic vulnerability analysis using a Geographic Information System, the gradual formation of clusters (for different values of k) allows a decisionmaking stakeholder to visualize more clearly the details of the cluster areas. The performance of the k-nearest neighbor graph method is tested on three classification problems, and finally it is applied to a sample from a digital map of Iaoi, a large city located in the North-Eastern part of Romania.