Attribute2oriented induction (AOI) approach is becoming one of the main techniques in data mining. In this paper , the basic AOI algorithm is described and its shortage is given. Then a new concept hierarchy optimiza2 tion method is presented. The rule2based concept graph is covered into concept tree ; the unbalanced tree ; is then covered into a balanced tree if concept tree is an unbalanced one ; the induction path of each node in the concept tree is recorded as node aggregate. The learning results are evaluated with support , confidence and LS sufficient fac2 tor. The algorithm is applied to Irish Educational Transitions Data , and results show that it is more effective and more suitable