Μελέτη και ανάπτυξη προσεγγίσεων αξιολόγησης αποτελεσμάτων συσταδοποίησης σε γράφους
View/
Keywords
Cluster validity ; Graph clustering ; Validity indicesAbstract
In this work we study the problem of cluster validity for data with graph structure. The problem of cluster validity has been studied widely and there are enough indices to evaluating different clustering results. There is a number of indices that measure the compactness and the separability of clustering results. However enough projects that have also studied these indices, cannot make clear conclusions about a set of different clustering results.
So with this work we study the problem of graph clustering and the problem of cluster validity too. On the second part we focus on the clustering validation techniques and presented various methods with different validation concepts.
At the rest of work (third and fourth part) we present an experimental approach where we choose different datasets with graph structure that clustered within different algorithms and for different cost functions. Alongside we evaluate these clustering results with three different clustering validation indices. At the end we compare all the indices with each other and choose the one that worked correctly based on his definition.