A local approach to rule induction in multi-scale decision tables

In order to represent hierarchical structure of data measured at different levels of granularities, the notion of multi-scale information table has been developed from the perspective of granular computation. In the present study, we mainly address the issue of rule induction in multi-scale decision tables. By considering independently three different types of complete multi-scale decision tables, we propose a local approach to rule induction. Compared with the existing literature, the selection of the optical level of scale and attribute reduction are both performed in a pointwise manner instead of the global one. Precisely speaking, the optical level of scale of any element relative to the decision attribute is firstly defined, then by removing superfluous attribute–value pairs in each decision rule, a set of simplified rules is derived. Lastly, a rule reduction is defined and the final set of decision rules is thus obtained.