A rough sets approach named semi-supervised learning  rough set(SLRS) is proposed based on semi-supervised learning. Some extended definitions of information theory, such as conditional mutual information between relative values are used to
measure the  attribute importance and relevancy in a single sample. And then attribute values are reduced heuristically by applying semi-supervised learning. The reduced decision table can be obtained. Even when processing incomplete information systems or lacking  prior knowledge, the proposed rules can be learned and added to the knowledge base. The experimental results on
UCI machine learning data sets and analysis of the instances show the reasonability and effectiveness of the proposed algorithm.