Discovering conjecturable rules through tree-based clustering analysis

We present a clustering technique to discover conjecturable rules from those datasets which do not have any predefined label class. The technique uses different attributes for clustering objects and building clustering trees. The similarity between objects will be determined using k-nearest neighbors graph, which allows both numerical and categorical attributes. The technique covers the convenience of unsupervised learning as well as the ability of prediction of decision trees. The technique is an unsupervised learning, making up of two steps: (a) constructing k-nearest neighbors graph; (b) building the clustering tree (Clus-Tree). We illustrate the use of our algorithm with an example.