To support the effective reasoning in the circumstances of incomplete andinsufficient facts, the author presented an inductive machine learning method, and designed a rule vector projection algorithm. With the method implemented by the algorithmthe primitive knowledge are processed by inductive approach, created a network knowledge base full of new classification concepts and paths for reasoning. Based on the knowledge base the quantity of facts needed in a diagnostic decision making by a machine inference system can sharply be reduced, thus a high performance can still be achieved by thesystem, despite of short information.